ikawrakow asked the seq_rm zeroing be hoisted out of the per-cell loop and
preferably done as part of the compute graph. This rewrites the fix:
- llama_kv_cache gets a pending_recurrent_reset bitmap, sized to
qnext_state_slots and indexed by slot index.
- llama_kv_cache_seq_rm only marks the bitmap; no CPU-side tensor
writes, no work inside the per-cell loop.
- delta_net::build_layer_attn_linear ORs the bitmap into the existing
reset_state flag, so the in-graph ggml_scale(state, 0.0f) path that
already handled batch.pos[0] == 0 now also handles slot reset.
- build_qwen3next and build_qwen35 clear the bitmap after
ggml_build_forward_expand, once per graph build.
- llama_kv_cache_clear also drops the bitmap, since the full-buffer
clear it already does makes any pending in-graph reset redundant.
Previous behavior is preserved (every recurrent layer column for the
released slot is zeroed before the next request reuses it), but the work
moves from the CPU into the compute graph, runs once per graph instead
of once per cell, and reuses the existing reset path referenced in the
review.
* server: spec checkpoints for recurrent models
* fix: save/restore sampler state during speculative checkpoint
When speculative decoding rejects draft tokens and restores the
recurrent state checkpoint, the sampler (RNG, grammar, prev tokens)
must also be restored to maintain consistency. Without this, the
sampler state reflects the rejected draft tokens, leading to
potential divergence.
Uses common_sampler_clone() to snapshot the sampler before the
speculative batch decode, and restores it on rejection.
* server: snapshot recurrent state in tensor
* reset ngram mod state for rejected tokens
* server: refactor checkpoint state logic
* speculative: fix sampler for checkpoints
* recurrent model: implement recurrent kernel checkpoint
* recurrent model: refactor api
* spec: free rbudget before overwriting
Two related issues that manifest as 'llama_decode ret=-3' on hybrid
architectures (e.g. Qwen3.5/3.6 MoE, Qwen3-Next), matching the symptom
reported in #1576.
1) server_context::apply_checkpoint() was written around transformer KV
semantics (pos_min / pos_max per-token window). For hybrid and pure
recurrent models the per-token pos_min threshold does not apply: the
recurrent state is a single snapshot, and the server-side checkpoint
is a whole-prefix record. The old selector 'cur.pos_min < pos_min_thold'
can succeed on a checkpoint whose pos_max is past the current n_past,
and — more commonly — fall through to do_reset = true, which zeros
slot.n_past / slot.n_past_prompt. Zeroing in-place while the recurrent
state in the context is still populated makes the next decode batch
disagree with the live state, returning ret=-3.
This change gates the checkpoint path on
llama_model_has_recurrent(llama_get_model(slot.ctx)):
- selector uses pos_max <= slot.n_past && pos_max < pos_next
(whole-prefix match, leaves at least one token to decode);
- on miss, slot state is preserved rather than zeroed, letting
update_slots() continue from the already-valid n_past_prompt;
- the erase loop drops any checkpoint whose pos_max > pos_next,
matching the rewind semantics for recurrent state.
Transformer behavior is unchanged.
2) stop_internal_decode is a file-static global in src/llama.cpp, set by
llama_decode_stop() (called on client disconnect) and polled inside
the decode loop to bail out with ret=-3. The flag is only cleared on
one conditional path in server_slot::release(), so a stop signal that
arrives after the interrupted llama_decode() has already returned
bleeds into the NEXT decode call and causes an immediate ret=-3 with
no work performed. Clear it at the top of the public llama_decode()
entry so the signal is scoped to the in-flight decode it was meant
for.
Build-verified: llama-server with GGML_CUDA=ON, -DCMAKE_CUDA_ARCHITECTURES=86
(sm_86), IQK flash-attn + matmul enabled. No new APIs introduced —
llama_model_has_recurrent is already public and already used elsewhere in
server-context.cpp.
Closes#1576
* Autoparser - complete refactoring of parser architecture
Autoparser: add optional argument reshuffle capability
Autoparser: True streaming (#20177)
* Relax atomicity constraint for nicer, more pleasent, True Streaming parsing
* Whitespace
* Remove redundant atomics
Revert to OAI-compatible args (#20213)
* Revert to OAI-compatible args
* Apply workaround::func_args_not_string
Fix structured outputs (#20223)
* Fix structured outputs
* Update common/chat-auto-parser-generator.cpp
Co-authored-by: Aldehir Rojas <hello@alde.dev>
---------
Co-authored-by: Aldehir Rojas <hello@alde.dev>
Fix compile bug (#20203)
* Fix compile bug
* Update common/chat-auto-parser-helpers.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
# Conflicts:
# common/chat-auto-parser-helpers.cpp
common : gracefully handle incomplete output (#20191)
* common : handle incomplete UTF-8 at end of input in PEG parser
* cont : if reached end prematurely, emit needs_more_input to propagate partial output
* cont: refactor peg parse context to add lenient flag
* cont : remove partial flag, keep lenient flag
PEG parser for LFM2 (#20251)
* PEG parser for LFM2
* Simplify using python_value()
common: map developer role to system (#20215)
* Map developer role to system
* Simplify
common: consolidate PEG string parsers (#20263)
* common : consolidate PEG string parsers
* cont : fix json_string_content()
examples : fix empty items in json_schema_to_grammar.py [no ci] (#19968)
* Fix logic for retrieving schema items in `json_schema_to_grammar.py`
If `schema['items']` is `{}` and `prefixItems not in schema', as `{}` is Falsy, the original code here will raise an error.
I think if `schema['items']` is `{}`, them items should just be `{}`
* Apply suggestion from @CISC
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Add tests for arrays with empty items
Add two unit tests to `tests/test-json-schema-to-grammar.cpp` that validate handling of arrays when 'items' is an empty schema and when 'prefixItems' is present alongside an empty 'items'. Both tests expect the same generated grammar, ensuring the JSON Schema->grammar conversion treats an empty 'items' schema (and the presence of 'prefixItems') correctly and covering this edge case.
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Reduce level of content parser warning message to avoid log spam on non-debug verbosity (#20347)
do not return if template parse failed
add arg to enable parallel tool call
common : fix incorrect uses of stoul (#20313)
# Conflicts:
# common/arg.cpp
# src/llama-grammar.cpp
examples : fix empty items in json_schema_to_grammar.py [no ci] (#19968)
* Fix logic for retrieving schema items in `json_schema_to_grammar.py`
If `schema['items']` is `{}` and `prefixItems not in schema', as `{}` is Falsy, the original code here will raise an error.
I think if `schema['items']` is `{}`, them items should just be `{}`
* Apply suggestion from @CISC
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Add tests for arrays with empty items
Add two unit tests to `tests/test-json-schema-to-grammar.cpp` that validate handling of arrays when 'items' is an empty schema and when 'prefixItems' is present alongside an empty 'items'. Both tests expect the same generated grammar, ensuring the JSON Schema->grammar conversion treats an empty 'items' schema (and the presence of 'prefixItems') correctly and covering this edge case.
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Add support for MiroThinker with new jinja template
common/parser: handle reasoning budget (#20297)
* v1
* Finished!
* Handlie cli
* Reasoning sampler
* Apply suggestions from code review
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Less explosive terminology :)
* Add utf-8 case and tests
* common : migrate reasoning budget sampler to common
* cont : clean up
* cont : expose state and allow passing as initial state
* cont : remove unused imports
* cont : update state machine doc string
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Co-authored-by: Alde Rojas <hello@alde.dev>
common/parser: use nlohmann::ordered_json to preserve parameter order (#20385)
common/parser: add GigaChatV3/3.1 models support (#19931)
Co-authored-by: Mishusha <pmv26021975@gmail.com>
common/parser: gracefully handle undetected tool parser, print error message. (#20286)
fix: prevent nullptr dereference (#20552)
common : fix iterator::end() dereference (#20445)
# Conflicts:
# common/regex-partial.cpp
jinja : add capability check for object args (#20612)
common/parser: add `--skip-chat-parsing` to force a pure content parser. (#20289)
* Add `--force-pure-content` to force a pure content parser.
* Update common/arg.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
common : rework gpt-oss parser (#20393)
* common : rework gpt-oss parser
* cont : fix gpt-oss tests
* cont : add structured output test
* cont : rename final to final_msg
common : fix gpt-oss content removal (#20745)
common/parser: add proper reasoning tag prefill reading (#20424)
* Implement proper prefill extraction
* Refactor cli parameters, update docs, move reasoning budget sampler part to common/reasoning-budget.cpp
* Update tools/server/server-task.cpp
* refactor: move grammars to variant, remove grammar_external, handle exception internally
* Make code less C++y
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
chat : handle tool calls with no required args in TAG_WITH_TAGGED format (#20764)
* chat : handle tool calls with no required args in TAG_WITH_TAGGED format
* Update tests/test-chat.cpp [no ci]
Co-authored-by: Aldehir Rojas <hello@alde.dev>
---------
Co-authored-by: Piotr Wilkin (ilintar) <piotr.wilkin@syndatis.com>
Co-authored-by: Aldehir Rojas <hello@alde.dev>
common/parser : fix out_of_range crash in throw path (#20424 regression) (#20777)
* chat : fix out_of_range crash in throw path (#20424 regression)
#20424 introduced effective_input = generation_prompt + input, but the
throw path uses input.substr(result.end) where result.end is a position
within effective_input. Every thinking model with a non-empty
generation_prompt crashes with std::out_of_range instead of the intended
error message.
Test crashes on unpatched master, passes with fix:
cmake -B build -DLLAMA_BUILD_TESTS=ON -DLLAMA_BUILD_TOOLS=OFF
cmake --build build --target test-chat
./build/bin/test-chat
* Update test-chat.cpp
* Update test-chat.cpp
* Update test-chat.cpp
---------
Co-authored-by: Piotr Wilkin (ilintar) <piotr.wilkin@syndatis.com>
jinja : fix heap OOB read in value equality comparison (#20782)
Address GHSA-q9j6-4hhc-rq9p and GHSA-2q4c-9gq5-5vfp.
The three-iterator overload of std::equal in value_array_t::equivalent()
and value_object_t::equivalent() reads past the end of the shorter
container when comparing arrays or objects of different lengths.
Use the four-iterator overload (C++14) which checks both range lengths.
Found-by: Pwno
common : fix typo in debug log ('extracft' -> 'extract') (#20807)
common/parser: fix nasty bug causing subtle corruption of generation prompt (#20825)
jinja : refactor token advancement (#20864)
* refactor token advancement
* exercise sub-expressions
common/autoparser : detect reasoning markers when enable_thinking changes system prompt (#20859)
common : replace wrap_for_generation with a prefix convenience function and fix gpt-oss (#20912)
jinja: fix macro with kwargs (#20960)
* jinja: fix macro with kwargs
* Apply suggestions from code review
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* fix newline problem
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
common : inhibit lazy grammar sampler while reasoning is active (#20970)
* common : inhibit grammar while reasoning budget is active
* cont : update force_pos in accept
* cont : fix tests
* cont : tweak should apply logic
* cont : return early not using grammar sampler
* Add tests
* cont : prevent backend sampling when reasoning budget enabled
* cont : fix typo
---------
Co-authored-by: Piotr Wilkin <piotr.wilkin@syndatis.com>
# Conflicts:
# common/reasoning-budget.h
# common/sampling.cpp
# tools/cli/cli.cpp
# tools/server/server-common.cpp
# tools/server/server-task.cpp
common/parser: fix reasoning whitespace bugs + extra parser tests (#21085)
* fix whitespace reasoning issues + add reconstruction tests
* Proper fix
* fix Nemotron autoparser test expectations to include newline in marker
common : add reasoning_format = none support to gpt-oss (#21094)
common/json-schema: fix: handle non-capturing groups (?:...) in JSON schema pattern converter (#21124)
The regex-to-grammar converter in _visit_pattern() crashes with SIGSEGV
when a JSON schema "pattern" field contains a non-capturing group (?:...).
Root cause: when the parser sees '(' followed by '?', it pushes a warning
but does not advance past '?:'. The recursive transform() call then
interprets '?' as a quantifier and calls seq.back() on an empty vector,
causing undefined behavior.
This commonly occurs when serving OpenAI-compatible tool calls from
clients that include complex regex patterns in their JSON schemas (e.g.,
date validation patterns like ^(?:(?:\d\d[2468][048]|...)-02-29|...)$).
The fix:
- Skip '?:' after '(' to treat non-capturing groups as regular groups
- For unsupported syntax (?=, ?!, etc.), skip to matching ')' safely,
handling escaped characters to avoid miscounting parenthesis depth
- Adjust the ')' unbalanced-parentheses check using direct char
comparisons instead of substr
- Add test cases for non-capturing groups (C++ only, as the JS/Python
implementations do not yet support this syntax)
common/parser: fix handling of tool definition with missing properties key (#21128)
jinja : handle empty expressions correctly (#20913)
* Reject empty computed member expressions before returning slices[0] from parse_member_expression_arguments().
* Treat empty computed member expressions with Jinja2 undefined semantics
Treat empty computed member expressions like `a[]` as undefined instead of
raising a parser error, to match Jinja2 behavior.
- return a noop expression for empty computed member arguments
- return undefined when a computed member key evaluates to undefined
- add Jinja tests covering `a[]|default('fallback')` and `a[] is undefined`
* Handle undefined computed member properties
Move undefined-property handling to the common member access path, and add a test covering `a[undefined] is undefined`.
* Use default undefined value in member access
Initialize val and then return it when property is undefined.
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* empty statement parses to blank_expression instead of noop_statement
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
common : gpt-oss handle builtin and unsolicited tool calls (#21213)
fix: tool call parsing for LFM2 and LFM2.5 models (#21242)
* fix: tool call parsing for LFM2 and LFM2.5 models'
* refactor: add test / break out lfm2 and lfm2.5 parsing logic
# Conflicts:
# common/chat.cpp
Relax prefill parser to allow space. (#21240)
* Relax prefill parser to allow space.
* Move changes from prefix() to parser generation
* Only allow spaces if we're not having a pure content parser next
common : add commentary rules for gpt-oss-20b (#21286)
add reasoning budget
model, mtmd: fix gguf conversion for audio/vision mmproj (#21309)
* fix gguf conversion for audio/vision mmproj
* fix test
# Conflicts:
# convert_hf_to_gguf.py
# examples/eval-callback/eval-callback.cpp
# examples/mtmd/CMakeLists.txt
# examples/mtmd/clip-impl.h
# examples/mtmd/mtmd.cpp
# gguf-py/gguf/constants.py
# gguf-py/gguf/gguf_writer.py
# gguf-py/gguf/tensor_mapping.py
# src/CMakeLists.txt
# src/llama-arch.cpp
# src/llama-arch.h
# src/llama-model.cpp
# src/llama-model.h
# src/llama-vocab.cpp
# src/models/models.h
# tests/test-llama-archs.cpp
# tools/mtmd/clip-graph.h
# tools/mtmd/clip-model.h
# tools/mtmd/clip.cpp
# tools/mtmd/models/models.h
fix: gemma 4 template (#21326)
chat : avoid including json in chat.h (#21306)
jinja: coerce input for string-specific filters (#21370)
common : fix tool call type detection for nullable and enum schemas (#21327)
* common : fix tool call type detection for nullable and enum schemas
* common, tests : fix grammar delegation for nullable/enum schemas and add tests
Fix enum type inference to scan all enum values (not just index 0) so
schemas like {"enum": [0, "celsius"]} correctly detect string type.
Fix schema_delegates in peg-parser to handle nullable type arrays
(["string", "null"]) and typeless enum schemas in raw mode, allowing
the tagged parser to use raw text instead of JSON-formatted strings.
Add test cases for Qwen3-Coder (TAG_WITH_TAGGED format):
- nullable string ["string", "null"]
- nullable string with null first ["null", "string"]
- nullable integer ["integer", "null"]
- enum without explicit type key
common/parser: fix call ID detection (Mistral parser mostly) + atomicity for tag-json parsers (#21230)
* Fix call ID detection (Mistral parser mostly) + atomicity for tag-json parsers
* Rename
* Update common/chat-auto-parser-generator.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
common : add gemma 4 specialized parser (#21418)
* common : add gemma4 dedicated parser
* cont : add '<|tool_response>' as eog
* cont : emit JSON from Gemma4 tool call AST
* cont : more fixes
* cont : refactor convert function
* cont : refine rules and mapping
* cont : add more tests
* cont : clean up
* cont : remove autoparser gemma4 implementation
* cont : more cleanup
* cont : rename gemma4.jinja to match the others
* cont : add custom template to support interleaved thinking
* cont : preserve reasoning in model turns
* cont : fix initializer error
* cont : fix unused vars
* cont : fix accidental static
* cont : fix specialized_template signature
* fix extra semicolon
* remove debug line and extra space [no ci]
fix reasoning budget
parser: fix MiniMax handling (#21573)
jinja : support ensure_ascii=true, string repetition and int/float self-filtering (#21623)
* feat: jinja engine improvements for reka-edge
Port three Jinja engine improvements needed for the reka-edge model:
1. Python-style string repetition ("ab" * 3 → "ababab")
2. ensure_ascii=true support for tojson filter (escapes non-ASCII to \uXXXX)
3. int() builtin on value_int_t (identity, needed for Reka Edge template)
* fix: escape invalid utf8 bytes when ensure_ascii=true
The json_ensure_ascii_preserving_format function does not correctly
handle an edge case where if UTF-8 parsing fails, it adds the non-ascii
character back to the output as a raw byte.
This commit fixes that by adding the unicode standard replacement
character \\ufffd to the output instead. This is the standard behavior
for various programming languages like Python, Rust, Go, etc.
* chore: address PR comments
1. Add todo comment for supporting string repetition for array/tuples
2. Add support for float identity operation
3. Move invalid ascii test case to test_fuzzing
* chore: accept suggestion for common/jinja/value.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
common : simplify autoparser tagged parser rules (#21216)
* common : simplify autoparser tagged parser rules
* cont : remove upper limit on optional args
* cont : revert changes to parsing at the end
* cont : undo arbitrary ordering of optional args
* cont : fix uninitialized required parameters
* revert to simplify merge
* re-apply patches
* restore flexible optional arg ordering tests
common : fix ambiguous grammar rule in gemma4 (#21661)
* common : fix ambiguous grammar rule in gemma4
* cont : fix missing comma...
common : enable reasoning budget sampler for gemma4 (#21697)
* fix: enable reasoning budget sampler for gemma4
Add thinking_start_tag and thinking_end_tag to
common_chat_params_init_gemma4(). Without these, the reasoning
budget sampler never activates for gemma4.
Make the newline after "thought" optional in the PEG parser to
handle budget=0 (sampler forces end tag before the newline).
Add test case for empty thinking block.
Fixes#21487
* use p.space() instead of p.optional(p.literal("\n")) in gemma4 thought parser
common : better align to the updated official gemma4 template (#21704)
fix: Fix broken structured output when using $refs in json_schema (#21699)
chat: dedicated DeepSeek v3.2 parser + "official" template (#21785)
Hide render_message_to_json warning
common/gemma4 : handle parsing edge cases (#21760)
common: skip reasoning budget sampler when no budget is requested (#21870)
* common: skip reasoning budget sampler when no budget is requested
After I added thinking_start_tag / thinking_end_tag for gemma4 in #21697, the reasoning budget sampler gets unconditionally created even when no budget is configured (the default -1). The same applies to kimi_k2, lfm2, lfm2_5, and ministral_3 which also set these tags. The budget gets converted to INT_MAX, so the sampler never actually forces any tokens but still runs per-token checks (start tag matching in IDLE state, token-to-piece conversion + UTF-8 checks in COUNTING state).
More importantly, the mere existence of the sampler (non-null rbudget) disables backend sampling. Backend sampling lets the GPU select tokens directly, avoiding a full logits transfer from GPU to CPU every token. This could explain the 30% speed regression reported in #21784 (98 t/s to 70 t/s on Vulkan).
So I added a reasoning_budget_tokens >= 0 check to the sampler creation condition. When the budget is unlimited, the sampler is not created, backend sampling stays enabled, and no per-token overhead is added. When a budget is explicitly set (0, 128, 1024, etc.), the sampler is created and works as before.
* common: preserve rbudget when grammar is lazy
Following up on the review feedback on #21870: keep the reasoning budget sampler when grammar_lazy is true, so the thinking-block grammar suppression from #20970 still works when tools are in use. This way, we only skip the sampler when both no budget is set AND grammar is not lazy.
autoparser: support case of JSON_NATIVE with per-call markers (test case: Reka-Edge) (#21892)
* fix grammar
* fix add sampled token
---------
Co-authored-by: Piotr Wilkin (ilintar) <piotr.wilkin@syndatis.com>
Co-authored-by: firecoperana <firecoperana>
Upstream: llama.cpp#21428. GGUF stores bools as 1-byte int8_t but the
loader was casting through (const bool*) which has implementation-defined
sizeof. Use explicit int8_t cast with != 0 normalization.
Co-authored-by: Mark Alonzo <mark.alonzo@outlook.com>
* WIP: Gemma4 vision
Crashes on the GPU because of rms_norm requiring ne0 to be multiple
of warp_size.
Runs on the CPU, but produces garbage.
* Remove unnecessary assert in CUDA rms_norm
* GLU was not advertised as supported on CUDA
* Still not working
* This seems to work
In llama_kv_cache_init() call, params.type_v_first was incorrectly passed
twice instead of params.type_v_last.
This caused V cache in the last N layers to use type_v_first instead of type_v_last.
Fix: Replace second params.type_v_first with params.type_v_last.
* wip: separate llama_context for MTP with graph reuse
* wip: fix KV cache desync with separate MTP context
* refactor: remove dead mtp logic code, encapsulate KV mirroring
* mtp-context: derive args directly from the main model's context
* mtp: fix kv cache positions
* clean small comments
* minor refactor for context shift
* Use build_std_attention for Gemma4 when possible
It is possible for the 26b MoE and 31b dense models.
It is not possible for the E4B/E2B vaiants because they
don't have KV cache in each layer.
* Standardize Gemma4 dense ffn
* WIP: Gemma4 split mode graph
Runs but produces NaNs
* WIP: Gemma4 split mode graph
Runs but very high PPL. At least it is no longer NaN.
* WIP
* This works!
* Put attn_norm, attn_post_norm, ffn_norm, ffn_post_norm on all GPUs
* Fix crash when saving/loading KV cache
* WIP: split mode graph for Gemma4-MoE - crashes
* Split mode graph for Gemma4-MoE - this works
* Disable SWA optimization
Something goes wrong there
* Consolidate MoE and dense graph parallel
* Use build_std_attention for Gemma4 when possible
It is possible for the 26b MoE and 31b dense models.
It is not possible for the E4B/E2B vaiants because they
don't have KV cache in each layer.
* Standardize Gemma4 dense ffn
* WIP: Gemma4 split mode graph
Runs but produces NaNs
* WIP: Gemma4 split mode graph
Runs but very high PPL. At least it is no longer NaN.
* WIP
* This works!
* Put attn_norm, attn_post_norm, ffn_norm, ffn_post_norm on all GPUs
* Fix crash when saving/loading KV cache
* Little maintenance
* llama-quantize : Add the missing items in the help
* Add GGML_MAX_CONTEXTS define in the general cmakelist.txt
* Make the KV cache (CPU) based warnings clearer
* Correct placement of GGML_MAX_CONTEXTS definition
* Revert wrong indents
This reverts commit d0728cbb6c6d4d6d8dc13f062e542d232647a38d.
* Moving the GGML_MAX_CONTEXTS definition to src/CMakeLists.txt
* Update warning message for unsupported KV cache types
* forgotten antislash
* Gemma4: WIP
* Gemma4: WIP - runs with totally wrong results
* Gemma4: WIP - add CPU 512, 512 FA
* Gemma4: WIP
It gives a meaningful response in llama-cli, but PPL is still much too
high. Is this due to tokenizer issues?
* Gemma4: this works
I had forgotten the softcap on the final output.
* Remove log
* Gemma4: WIP E4B/E2B
* Gemma4: Q4B/E2B appear to work now
* gemma4: tokenizer fixes
* wip: port MTP architecture
Ports the Multi-Token Prediction (MTP) architecture to the older `llama.cpp` codebase used by `ikllama`.
Changes include:
- Updating `llama_batch` to support `mtp_params`.
- Modifying `llama_decode_internal` (and `encode`) to handle MTP operations (Warmup, Update, Draft).
- Adding public APIs for MTP state management (`llama_set_draft_input_hidden_state`).
- Adapting the embedding extraction logic to skip MTP update passes.
* Refactors `server_slot` to support generic speculative decoding (MTP or Draft Model).
* core: enable hybrid outputs (logits + embeddings) for MTP support
* fix(mtp): correct KV-cache slot finding for updates
* fix(mtp): persist hidden states to prevent context corruption during drafting
* refactor(mtp): clean unused code
* fix(mtp): update server to new functions name
* fix(mtp): fix graph and save hidden state
* mtp: refactor integration, context params and kv cache search
* mtp: fix hidden state extraction and speculative acceptance flow
* server: fix MTP warmup for long prompts and reset token buffer
* llama: refactor MTP operation state to context parameters
* server: fix n_past calculation in MTP acceptance
* llama: fix mtp enable flags
* speculative: refactor MTP to use common_speculative interface
* context: remove unused signatures
* clip: fix deprecated enum-enum conversion warning
* common: fix format string crash in help message
* context: fix mtp activation logic
* llamat: always use the extracted embedding
* llama: get all embeddings to kv cache
* llama: revert logit to not run mtp for not supported arch
* llama: allocate all the n_outputs for MTP
* wip
* server-context: get only the last embedding for hidden state
* ggml-backend: fix array of bounds in debug build
* server-context: run mt kv update to each prompt batch
* revert segmentation fault fixes
* glm-mtp(feat): optimize graph embedding and recursive drafting
* Allow using -rtr and -muge together
* Various Qwen-3.5 tweaks
* No need to make v, g, beta contiguous
* Adjust NEON delta-net to non-contiguous v, g, b
* Cleanup
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* simpler n_rewind logic, delete old comments
* use more consistent names, add updt_w_cur to json schema
* align comments
* refactor review logic, update struct/variable names
* revert cosmetic changes
* check enable/disable in llama_prep_adaptive_p_impl()
* delete extra whitespaces after statement
* show target in debug prints
* more concise debug print
* delete old comments
* update with loop instead of move()
* comment out all adaptive p debug prints
* more debug prints
* move review() variables: common_sampler struct -> common_sampler_review() args
* match n_unsent type
* fix merge bugs, delete adaptive p references in buffer_and_check_string_ban()
* restore accidental erasure
* Revert "adaptive p: collect probability before logit bias"
This reverts commit 1434878461c49d1a2a9047fc15d5e7b78421fd2a.
* server : support multi-modal context checkpoints and prompt caching
do not create checkpoint right after image processing
improve mtmd check for slot ops
fix context shift
do not abort if template parse failed
* change to debug message when detecting ban token
---------
Co-authored-by: firecoperana <firecoperana>
* WIP: support pre-merged up/gate experts
Haha, mainline has elected to arrange the merged tensors
the other way around compared to what I had done in the on-the-fly merge.
* Change the order of on-the-fly packed up/gate
* OpenAI
* CUDA TG
* CPU
* Split mode graph for models with pre-merged ffn_up/ffn_gate experts
* WIP: support pre-merged up/gate experts
Haha, mainline has elected to arrange the merged tensors
the other way around compared to what I had done in the on-the-fly merge.
* Change the order of on-the-fly packed up/gate
* OpenAI
* CUDA TG
* CPU
* WIP
* WIP
* WIP
* WIP
* WIP
* WIP
* WIP
Loads and starts running, crashes with illegal memory access in
quantize_mmq_q8_1. This almost always indicates NaNs in the input
to the MoE FFN part.
* WIP
* WIP
Loads and runs, wrong results (very high PPL)
Performance looks promising, around 25% better than previous sm graph.
Needs f32 or bf16 graph reduce type.
* WIP - still wrong
* Fix after rebase
* WIP
* WIP
* This seems to be working for dense Qwen3.5!!!
* WIP: Qwen3-Next is not quite working
* Some cleanup
* Disable Qwen3-Next for now
* Disable graph parallel when mmproj was specified
* Read/write split recurrent state
* That should not crash
* Re-enable vision - it works now
* Recurrent layers should now be counted for split cache
---------
Co-authored-by: Piotr Wilkin <piotr.wilkin@syndatis.com>
common : add nemotron 3 parsing (#18077)
common : add parser for ministral/mistral large 3/devstral 2 (#17713)
common : default content to an empty string (#18485)
chat: make tool description and parameters optional per OpenAI spec (#18478)
Per the OpenAI API specification, both 'description' and 'parameters'
fields in tool function definitions are optional. Previously, the parser
would throw an exception if these fields were missing.
Attempts to fix#17667
common : implement new jinja template engine (#18462)
---------
Co-authored-by: Alde Rojas <hello@alde.dev>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
jinja: correct member access rule (#18905)
jinja : fix lexing of float literals with sign (#18901)
jinja : add missing tojson filter for bool (#18900)
jinja : attribute support for join, map and sort (#18883)
jinja : fix object item order (and properly implement dictsort) (#18904)
tests : add test-jinja -py option for cross-checking (#18906)
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
ci : run test-jinja -py on high perf [no ci] (#18916)
jinja : fix undefined keys and attributes and int/float as bool (#18924)
jinja: support none|string (#18995)
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
jinja : implement mixed type object keys (#18955)
---------
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
jinja : undefined should be treated as sequence/iterable (return string/array) by filters/tests (#19147)
`tojson` is not a supported `undefined` filter
keep it DRY and fix some types
jinja : do not pass empty tools and add some none filters (#19176)
jinja : add unordered_map include to value.h [no ci] (#19205)
jinja : add missing 'in' test to template engine (#19004) (#19239)
The jinja template parser was missing the 'in' test from
global_builtins(), causing templates using reject("in", ...),
select("in", ...), or 'x is in(y)' to fail with
"selectattr: unknown test 'in'".
This broke tool-calling for Qwen3-Coder and any other model
whose chat template uses the 'in' test.
Added test_is_in supporting array, string, and object containment
checks, mirroring the existing 'in' operator logic in runtime.cpp.
Includes test cases for all three containment types plus
reject/select filter usage.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
---------
Co-authored-by: Sid Mohan <sidmohan0@users.noreply.github.com>
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
Add Jinja support for "indent" string filter (#19529)
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
add vendor
refactor chat
server : support preserving reasoning_content in assistant message (#18994)
chat : fix translategemma crash on common_chat_format_example (#19019)
chat: fix language input for translategemma (#19052)
Co-authored-by: Aldehir Rojas <hello@alde.dev>
---------
Co-authored-by: Aldehir Rojas <hello@alde.dev>
chat: fix case where template accepts type content only (#19419)
mtmd : chat : Fix extra \n between text and media marker (#19595)
Thanks to @tugot17 for detecting and reporting the issue.
For vision models (e.g. LFM2.5-VL-1.6B and Qwen/Qwen3-VL-4B-Instruct) `llama-mtmd-cli` produces identical output to HF implementation.
However `llama-server` doesn't. I traced it down to extra newline
inserted after `<__media__>`.
This happens in `to_json_oaicompat`, that treats media markers as text
and joins all parts with `\n` separator.
PR introduces new type `media_marker` and uses it for media markers.
Extra logic is added to prevent insertion of newlines before and after
media markers.
With this change number of input tokens is identical to HF
implementation and as a result the output is also identical.
I explored other ways to address the issue
* remove completely `\n` between text parts in `to_json_oaicompat`
* merge text messages in server-common.cpp before sending them to `to_json_oaicompat`
Please propose alternative ways of fixing this issue.
Co-authored-by: Piotr Wilkin (ilintar) <piotr.wilkin@syndatis.com>
---------
Co-authored-by: Piotr Wilkin (ilintar) <piotr.wilkin@syndatis.com>
common : merge qwen3-coder and nemotron nano 3 parsers (#19765)
common : fix improper trimming in XML parser on complete message (#19805)
Co-authored-by: Jules LEIDELINGER <11395311+julio75012@users.noreply.github.com>
jinja: correct stats for tojson and string filters (#19785)
jinja : correct default size for string slices (#19913)
common : handle unicode during partial json parsing (#16526)
common : fix json schema with '\' in literals (#17307)
add back qwen_coder_xml and mirothinker
Co-authored-by: Aldehir Rojas <hello@alde.dev>
* fix adaptive p sampler rewinding too far back
* update comments
* correct default value for total_weight, more comments
* new variables/names
* update comment for n_rewind
* move null pointer check back to common_sampler_review()
* refactor weighted_sum and total_weight to vector<pair>, better boundary check in llama_review_adaptive_p_impl()
* Revive fused delta-net
* Add command line argument for fused delta net
* Simplify/improve CUDA delta-net
* Add -fdn to llama-bench
* More CUDA fused delta net optimizations
* CPU optimizations
* Much faster fused delta-net on the CPU
It seems it is faster than the chunked implementation!
* Change meaning of fdn from bool flag to threshold value
* Use eps = 1e-6
* Give some nodes a name
* Don't re-apply L2 norm - it has already been done
* This seems quite a bit better
* More tweaks
* Restore per context buffer size log
Not everybody uses models split in 2000 parts, and those who do,
actually want to see the biffer sizes.
* server: enable checkpoint for recurrent models
create checkpoint after cancel
fix ban string and rm context during rewind
add checkpoint interval
only save recurrent cache
* save checkpoint during pp
---------
Co-authored-by: firecoperana <firecoperana>
* Revive fused delta-net
* Add command line argument for fused delta net
* Simplify/improve CUDA delta-net
* Add -fdn to llama-bench
* More CUDA fused delta net optimizations
* CPU optimizations
* Much faster fused delta-net on the CPU
It seems it is faster than the chunked implementation!
* Change meaning of fdn from bool flag to threshold value
* Use eps = 1e-6
* Give some nodes a name
* Display the size of the tensors overriden during the tensor loading
Ex:
`Tensor blk.60.ffn_gate_exps.weight buffer type overriden to CPU
Tensor blk.60.ffn_up_exps.weight buffer type overriden to CPU`
become
`Tensor blk.60.ffn_up_exps.weight (size = 668467200 bytes) buffer type overriden to CPU
Tensor blk.60.ffn_gate_exps.weight (size = 668467200 bytes) buffer type overriden to CPU`
And pass in debug the later displayed size of the unnamed buffer overrides.
Ex : `llm_load_tensors: CPU buffer size = XXX.XX MiB`
That double display is cluttering the screen without being very informative.
* change bytes display to MiB.
Co-authored-by: Kawrakow <iwankawrakow@gmail.com>
---------
Co-authored-by: Kawrakow <iwankawrakow@gmail.com>
* Partial Requant feature for llama-quantize
- Inspired by the recently portcopied --dry-run feature.
- Allows to partially requantize a split quantized .gguf by requantizing only the missing splits in the destination directory.
- Works both for GGUF which are split tensors by tensors, or by group of several tensors (though this one is not very much tested beyond 2 tensors by split).
- Vibe coded.
* Create output directory if it doesn't exist in llama-quantize
* Create output directory if it doesn't exist in gguf-split
* Add exit when directory fails to be created on Windows
* Use std::filesystem
* cleanup
* wip: port MTP architecture
Ports the Multi-Token Prediction (MTP) architecture to the older `llama.cpp` codebase used by `ikllama`.
Changes include:
- Updating `llama_batch` to support `mtp_params`.
- Modifying `llama_decode_internal` (and `encode`) to handle MTP operations (Warmup, Update, Draft).
- Adding public APIs for MTP state management (`llama_set_draft_input_hidden_state`).
- Adapting the embedding extraction logic to skip MTP update passes.
* Refactors `server_slot` to support generic speculative decoding (MTP or Draft Model).
* core: enable hybrid outputs (logits + embeddings) for MTP support
* fix(mtp): correct KV-cache slot finding for updates
* fix(mtp): persist hidden states to prevent context corruption during drafting
* refactor(mtp): clean unused code
* fix(mtp): update server to new functions name
* fix(mtp): fix graph and save hidden state
* mtp: refactor integration, context params and kv cache search
* mtp: fix hidden state extraction and speculative acceptance flow
* server: fix MTP warmup for long prompts and reset token buffer
* llama: refactor MTP operation state to context parameters
* server: fix n_past calculation in MTP acceptance
* llama: fix mtp enable flags
* speculative: refactor MTP to use common_speculative interface
* context: remove unused signatures
* clip: fix deprecated enum-enum conversion warning
* common: fix format string crash in help message
* context: fix mtp activation logic
* adaptive p: upadte internal state only if not rewinding
* adaptive p: conditional update for speculative decoding
* adaptive p: refactor to rewind instead of update
* adaptive p fix: better comments
* fix rewind check
* add record to handle multi-token rewind
* better comment
* Optimizing q3next TG
* Fused add -> softplus -> mul on CUDA
* Remove forgotten debug log
* Increase ggml context size
Required for Qwen3-Next with batch/u-batch size of 4096
* WIP
* Avoid some contiguous ops
* Avoid some repeats
* Avoid some more repeats
* qwen3next: add architecture support and recurrent-state fixes
* qwen3next: optimize broadcast sub and single-seq ssm conv
* cuda: build MoE row mapping on device in mul_mat_id
* cuda: add guarded multi-seq fast path for ssm_conv
* docs: update qwen3next perf report for cuda MoE/SSM tuning
* cuda: reduce qwen3next moe/ssm sync overhead and refresh eval
* qwen3next: split cpu/cuda eval builds and tune PP scheduling
* qwen3next: harden seq-state flow and support optional dense FFN layers
* qwen3next: trim delta-net graph overhead in chunking path
* qwen3next: remove redundant v_conv cont in delta path
* qwen3next: avoid extra cont on linear attention output
* qwen3next: drop redundant cont before recurrent state flatten
* qwen3next: keep recurrent state in 4d layout through delta path
* qwen3next: add fused delta-net op and wire model path
* tests: add backend-op coverage for ggml_delta_net
* qwen3next: add runtime switch for fused delta-net path
* docs: refresh qwen3next perf review and benchmark matrix
* qwen3next: default fused delta-net off and document quality checks
* qwen3next: add decode-only fused delta mode
* qwen3next: make fused delta safe by default and fix fused tensor layout
* qwen3next: warn when forcing fused decode mode
* qwen3next: add fused-delta regression runner script
* qwen3next: integrate fused regression into eval harness
* qwen3next: clean up chunked delta-net shape handling
* qwen3next: add absolute sanity guards to fused regression
* qwen3next: add unified regression runner script
* qwen3next: disable flash-attn for cpu-only contexts
* docs: reconcile qwen3next status and remaining upstream gaps
* common: add qwen3next fused-delta runtime flag
* cuda: add qwen3next delta-net kernel dispatch override
* docs: update qwen3next quality and serving baseline findings
* qwen3next: keep fused delta on safe path and remove PR artifacts
* qwen3next: align autoregressive delta-net decode layout
* Revert "qwen3next: align autoregressive delta-net decode layout"
This reverts commit 9241164a5ea9e032a2456fbf2dd0bf798b264fd7.
* cuda: port solve-tri fast-paths for qwen3next delta-net
* qwen3next: add fused-delta runtime flag and drop env toggle
* qwen3next: make fused delta single-flag and default on
* Account for GPU arch differences
* Revert "cuda: build MoE row mapping on device in mul_mat_id"
This reverts commit 89e9ecfa840b04e88699ab3803eb732cd78727f9.
* qwen3next: drop non-essential MoE scheduling and split heuristics
* qwen3next: avoid generic ggml_sub broadcast changes
* llama: restore only_active_experts log message
* Remove unnecessary hacks, disable fusion for now.
* qwen3next: port hybrid recurrent state memory semantics
* qwen3next: clean up recurrent state slot plumbing
* qwen3next: fix hybrid V-cache layout plumbing
* qwen3next: guard recurrent state slots against kv capacity
* qwen3next: persist recurrent state in session data
- serialize/restore qwen3next cache.s_l in state/session paths\n- bump session and sequence-state file versions for format change\n- fallback to single-token chunking for mixed repeated seq_id batches
* qwen3next: drop unused fused-delta builder path
- remove dead build_delta_net_fused lambda\n- remove unused llm_build_context::fused_delta member
* qwen3next: remove unused fused-delta CLI/context plumbing
- drop -fd/-no-fd options and related YAML dump field\n- remove fused_delta fields from public/internal context params\n- remove fused_delta assignment and logging in context init
* ggml: remove unused DELTA_NET operator stack
* Missing include
* Reorder ops/unary ops
So we don't change again the enum values of the mul mat ops
* Minor
* Discard unnecessary changes in llama-build-context.cpp
* Minor
* Revert "Discard unnecessary changes in llama-build-context.cpp"
This reverts commit edadb80ed68c4c0831e9c22609a9a3af19be9735.
* Increase GGML_SCHED_MAX_SPLITS - required for larger u-batches
* Fix CPU concat in the TG case: 7.25 -> 10.5 t/s for Qwen3Next
* Fix CPU sum_rows: 10.5 -> 13.6 t/s for Qwen3Next
It was single-threaded and was taking ~25% of the computation time
during TG. It is now down to 2%.
Strangely enough, I measure 13.6 t/s with llama-bench, but if I
let the model give me an actual response with llama-cli, I get close
to 17 t/s.
* Fix CPU scale: 13.6 -> 16.7 t/s for Qwen3Next
For Qwen3Next there is a scale op on a largish tensor (548k elements)
that has a single row for TG, so was done in a single thread.
We now simply use blocks of 1024 elements.
* Optimize CPU mul: 16.7 -> 17.6 t/s for Qwen3Next
* CPU: fuse transpose -> cont -> sum_rows -> transpos: 17.6 -> 23.1 t/s for Qwen3Next
* Optimize CPU repeat: 176 -> 200 t/s for Qwen3Next PP-512
* Multithreading for OP_SUB
* Don't commit with timing trace on
* Multithread neg and sigmoid
* Be able to turn on/off fusion more easily (CPU)
* Name the mul_mat ops so we know where the time goes
* WIP
* Much better PP on CUDA
* CUDA: fuse transpose -> cont -> sum_rows -> transpose
Needs non-coontiguous variant of sum_rows.
On the CPU this gave 30+% improvement in TG performance,
on CUDA ist is disapointing 6-7%. I guess, this is because
Georgi's cont CPU implementation was so bad that skipping
it made such a big difference.
* CUDA: faster mul for special case relevant for Qwen3Next
Worth 1% in TG
* Fix CPU OP_CONT
---------
Co-authored-by: yurko <yurko@local>
Co-authored-by: Yurko <yurko@example.com>
Co-authored-by: yurko <yurko@pop-os.tail5a1a6b.ts.net>
Co-authored-by: Yurko Hoshko <YurkoHoshko@users.noreply.github.com>
* spec : add self speculative decoding and ngram-mod and refactor
common : use common_ prefix for common library function
llama : use LLAMA_TOKEN_NULL
spec : add self speculative decoding (no draft model required) + refactor
spec : add ngram-mod
spec : various improvements ton ngram-map + docs
spec : fix the check-rate logic of ngram-simple
common : add common_speculative_is_compat()
spec : simplify time measurement using common_time_meas
refactor common_sampler_init
refactor common_token_to_piece
refactor and fix cur_p bug
clean up
* spec : remove check rate
* spec: show warnings instead of abort
---------
Co-authored-by: firecoperana <firecoperana>
Co-authored-by: Sascha Rogmann <59577610+srogmann@users.noreply.github.com>
* WIP
* This works but is slow
* Turn off the up / gate clamps for now
* OK we need the clamping
* Fuse the clamp (CUDA)
* Fuse the clamp (CPU)
* WIP
* Be able to use merged q, k, v
* Be able to use merged up/gate experts
* Fuse the clamp (CUDA mmvq)
* WIP: graph parallel for Step-3.5
* WIP
* This should be it
* Cleanup
* Fix merge
* Not working attempt to extend fused_mul_unary to the Step-3.5 case
* It works now, but performance gain is very minor
* Fix graph parallel when ngl < n_layers
* Fix using ffn_norm
When using graph parallel with ngl < n_layers, the ffn_norm tensor
may have ended up being split, while the ffn tensors are on the CPU.
In that case we will get a crash because we attempt to use the not-split
buffer of ffn_norm, which is invalid. Thi commit fixes that.
* Cleanup
* WIP
* This works but is slow
* Turn off the up / gate clamps for now
* OK we need the clamping
* Fuse the clamp (CUDA)
* Fuse the clamp (CPU)
* WIP
* Be able to use merged q, k, v
* Be able to use merged up/gate experts
* Fuse the clamp (CUDA mmvq)
* WIP: graph parallel for Step-3.5
* WIP
* This should be it
* Cleanup
* Fix merge
* WIP
* This works but is slow
* Turn off the up / gate clamps for now
* OK we need the clamping
* Fuse the clamp (CUDA)
* Fuse the clamp (CPU)
* WIP
* Be able to use merged q, k, v
* Be able to use merged up/gate experts
* Fuse the clamp (CUDA mmvq)
* adaptive_p: fix history update + use current probability for high temp
* adaptive_p: fix history update bug, update with current probability if temp is high
* replace temp-as-signal with server argument
* adaptive_p: rename ema_w_cur_p to updt_w_cur
* delete test code
* Copy reduce result to other GPUs if necessary
* Avoid ggml_get_rows for TG
* For the output ops use the result of the split that ran on the main GPU
* More models
* A hopefully more efficient adaptive_p sampling
* Once at it, lets fix the formatting too
* More formatting
* Hopefully better
* This should be better
* Correctly accumulate adaptive_p sampling time
* AVX2
* A hopefully more efficient adaptive_p sampling
* Once at it, lets fix the formatting too
* More formatting
* Correctly accumulate sampling time for adaptive_p
* adaptive-p sampler: fix zeroed orig_probs bug and refactor
- Fix bug where original probabilities were captured as zero by calculating
them from logits in llama_prep_adaptive_p (new).
- Replace vector with unordered_map to track candidate probabilities,
filtering for relevance via logit delta (16.6f).
- Standardize API naming: llama_<action/verb>_<focus/name/topic>_<extra/info>
- Update function signatures to follow most other samplers.
* resolve merge bug
* adaptive-p: revert reordering function definitions
* Attempt to fix the many GPU issue in split mode graph
* WIP: this seems more stable
Still hanging after a while if I try to use all 7 GPUs
* Reenable OpenMP in scheduler async
Seems solid up to 4 GPUs. It did hang with --max-gpu 6.
* printf cleanup
* Add ability to merge up+gate exps to more models
* We need to of course pass the merged tensor to build_ffn
* All the others
* Also Qwen3VL-MoE
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* WIP - not working
* WIP - not working
* WIP - GPT-OSS working
However, extremely stupid. The only way I could correctly repack the
up/gate experts is to copy up and gate into host buffers, repack
into another host buffer, copy back into the ffn_up_gate_exps tensor.
This is going to be very slow for giant 500 GB models.
My attempts to do this via a compute graph on the backend holding
the tensors was unsuccessful.
For GPT-OSS-20B I see ~6-7% better PP when using the original
ik_llama.cpp fused_up_gate CUDA implementation, and ~10% when
using the small batch size implementation.
Other models are not working yet on CUDA as I need to fix the
fused mul-unary implementation.
* WIP
* WIP - Qwen3-MoE (and hopefully all others) working
But when I say here and in the previous commit "working",
I mean PP is working. TG is still broken.
* WIP: TG seems to be working
* Minor
* Add command line option to merge experts up/gate
* Add merge up/gate command line parameter to llama-bench
* Turn off merge_up_gate_exps if split mode graph
It is not yet implemented
* When no bias, allow merging up/gate with tensor overrides
* Arghh, we need to increase the context size again
* Cleanup
* Better VRAM utilization strategy for split mode graph
* Fix assert when --max-gpu is less than available GPUs
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* Mimo-2 support
* Fix bug for head sizes not being the same
It still does not solve the Mimo-2 quantized cache issue.
* Fix quantized cache
* Minor
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* WIP: absorb adding input into std_attn and std_ffn
* WIP: NCCL infra
* WIP: add reduce and fake_cpy ops
* WIP
* WIP: graph appears to work, layer is broken
* WIP: Qwen3-MoE works with graph, layer still broken
* WIP: GLM-4.5 graph works
* WIP: fix sm layer (dense)
* WIP: fix sm layer (MoE)
* WIP: fast PP with bespoke 4-GPU NCCL
I guess, I'm not using NCCL the right way as PP is very
low with a single communicator group for 3 or more GPUs.
But if I create 4 communicator groups for pairs of GPUs
(0,1, 2,3, 0,2, 1,3) and use that, PP is fast: I'm hitting
1500 t/s for L3-70B on the 4x3090 system, which is
~20% better than the previous sm graph without NCCL.
But that cannot be the solution (I cannot be creating pairwise
communicators and associated logic for every possible number of GPUs).
* WIP: Cohere2
* Explicitely set device
* Bespoke 3-GPU case
* WIP
* Do not repeat get_rows multiple times
* Fix 3 GPUs
* OK, let's leave it in
* Simple async
* This sync seems enough
* Only do async for 4 or more backends
With 2 GPUs (so, 3 backends) not using async is slightly faster
* Scheduler changes
* Use OpenMP if available
Surprisingly (at least to me), this is quite a bit faster than
std::thread and std::barrier. GLM-4.5-AIR with 4 GPUs is now
at 105 t/s at zero context!
* Do not use OpenMP if there are tensor overrides
* Set omp max active levels
* Be more careful with having set the device before using a stream
* Command line option to turn on async. Set to false by defualt for now
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* Adding fused_norm - same idea as fused_rms_norm
* Avoid computing the attention reduce op for cohere2
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* WIP: absorb adding input into std_attn and std_ffn
* WIP: NCCL infra
* WIP: add reduce and fake_cpy ops
* WIP
* WIP: graph appears to work, layer is broken
* WIP: Qwen3-MoE works with graph, layer still broken
* WIP: GLM-4.5 graph works
* WIP: fix sm layer (dense)
* WIP: fix sm layer (MoE)
* WIP: fast PP with bespoke 4-GPU NCCL
I guess, I'm not using NCCL the right way as PP is very
low with a single communicator group for 3 or more GPUs.
But if I create 4 communicator groups for pairs of GPUs
(0,1, 2,3, 0,2, 1,3) and use that, PP is fast: I'm hitting
1500 t/s for L3-70B on the 4x3090 system, which is
~20% better than the previous sm graph without NCCL.
But that cannot be the solution (I cannot be creating pairwise
communicators and associated logic for every possible number of GPUs).
* WIP: Cohere2
* Explicitely set device
* Bespoke 3-GPU case
* WIP
* Do not repeat get_rows multiple times
* Fix 3 GPUs
* OK, let's leave it in
* Implement the reduce op without NCCL available
* Be able to build without NCCL
cmake -DGGML_NCCL=OFF disables it
* Make --max-gpu work again
* Slightly better for 4 GPUs without NCCL
* Cleanup
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Use -smgs or --split-mode-graph-scheduling in CLI to bypass the disabling of split mode graph scheduling when tensor overrides is used.
Co-authored-by: Kawrakow <iwankawrakow@gmail.com>
* This should do the trick for PP
* Command line option to set max. extra VRAM that the scheduler can use
* Fix bug and cleanup
* Looks like with this change it is working with tensor overrides
* Nah, it is not working
* OK, this seems to be working
* Disable split scheduling with tensor overrides
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* This works and TG is descent, but PP is low
* Better
* Apply f_logit_scale before mul mat with output tensor
* This is better for PP: 600 t/s -> 700 t/s
* To not lose this again
* WIP
* Equal split
* WIP
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* Rearrange graph nodes
So that we can do graph portions that are the same on 2 or more
GPUs at the same time.
* Separate graph compute implementation for split mode graph
* This is better
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Required adding the "temperature scaling" to the standard attention
implementation.
But in this way split mode "graph" is automatically supported.
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* Remove most of split mode row
* WIP
* WIP: also allocate the KV cache using tensor split
* WIP: it runs with wrong result
But it also looks like the backend scheduler is not going to help:
* It copies mask and input positions to GPU 0
* => RoPE ops must run on GPU 0
* => To proceed attn evaluation, GPU 1 must wait for GPU 0 to finish its
entire attn calculation
* Same with FFN. The rms_norm gets scheduled on GPU 0. Hence, GPU 1 must
wait for GPU 0 to finish its entore FFN calculation before it can
start (as it needs to copy the result of rms_norm from GPU 0)
* => Seems useless without writing a bespoke TP scheduling
* WIP
* This works, but it is slow
* This is slightly better
the graph is still not being computed in parallel.
Why? Because the scheduler creates graph splits where the
result of the computation on one GPU becomes an input for the
other split. Hence, to trigger the computation on the second GPU
one needs to wait for the computation on the first GPU to finish,
even thiough the two can be done in parallel up to the sunchronization
point. So, all that is left to do is to trick the scheduler to create
to splits that can be done in parallel, and then have a graph split
where the results get combined.
* Playing games with the scheduler
This change tricks it into doing the right thing^TM.
Still quite a bit slower than split mode layer for the 8B LlaMA model.
But for the 70B LlaMA it now beats split mode layer for TG:
28 t/s vs 24.4 t/s. PP is 627 t/s vs 744 t/s.
In comparison, split mode "row" in mainline gets
484 t/s PP and 19.3 t/s TG.
* Fix attn split
Granularity for Wq, Wo is not just head size, but
head size * gqa_ratio.
Else the Wk, Wv tensors end up not being a multiple of the
head size when we divide the split determined by Wo with
the gqa_ratio.
* Show memory used per device
* Make it work with partial offload
but no tensor overrides yet, just ngl < num_layers.
* Allow for f16 source in fused_rms_norm
* This results in faster PP.
Now PP is faster than split mode layer for L3-70B.
* Rename split mode "row" to split mode "graph"
* Leave FFN partial results as f16
* WIP GLM4.5 - runs with wrong results
* WIP GLM4.5 - this works
PP is already better than split mode layer, but TG for zero context
is kind of low - 60 vs 92 t/s. TG becomes better than split mode layer
at around 20k tokens. PP at 26k tokens is 1.55X of sm layer.
* Work around compiler bug
It issues a warning that there is an extra semicolon outside of a function,
but there isn't. If I remove the anonymous namespace and turn the
functions inside into static, the warning disapears, so clearly
a compiler bug.
* Make graph reuse work with split mode graph
* Remove more split mode row remnants
* WIP tensor overrides
Runs with wrong results, don't see where the issue could be.
* This works but is slow
Still does not work for row-interleaved quants
* Slightly better
* Slightly better
* Row-interleaved quants work
* Better
* Minor
* Guarad against using split mode "graph" for unsupported models
* Guards against using merge_qkv with split mode "graph"
* WIP split mode attn
Works for LlaMA models, but not for GLM-4.5.
Doesn't seem to improve performance, so I guess no point in trying to
fix it.
* Split mode graph for qwen3moe
* Try to better distribute the splits
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* RPC support multiple devices
* rpc : update documentation (#16441)
Update the README file to match the newly added functionality of
exposing multiple devices from a single server.
Co-authored-by: Diego Devesa <slarengh@gmail.com>
# Conflicts:
# examples/rpc/README.md
* Remove memory settings
* rpc : cache and reuse compute graphs (#15405)
Store the last computed graph and reuse it when possible.
Also do not return response from GRAPH_COMPUTE and assume it always
completes successfully. If this this is not the case, the server closes
the connection. This saves us a network round trip to the server.
* Add -cpu to include cpu backend
---------
Co-authored-by: firecoperana <firecoperana>
Co-authored-by: Radoslav Gerganov <rgerganov@gmail.com>
* grammar : fix JSON Schema for string regex with top-level alt. (#9903)
Prior to this commit, using a JSON Schema containing a string
with `pattern` regular expression that uses top-level alternation
(e.g. `"pattern": "^A|B|C|D$"`) would result in invalid JSON
output from the constrained sampling grammar, because it
ended up creating a grammar rule like this for the string:
```
thing ::= "\"" "A" | "B" | "C" | "D" "\"" space
```
Note that this rule will only match a starting quote for the "A" case,
and will only match an ending quote for the "D" case,
so this rule will always produce invalid JSON when used for sampling
(that is, the JSON will always be lacking the starting quote,
the ending quote, or both).
This was fixed in a simple way by adding parentheses to the
generated rule (for all string pattern rules, to keep it simple),
such that the new generated rule looks like this (correct):
```
thing ::= "\"" ("A" | "B" | "C" | "D") "\"" space
```
* grammars : add English-only grammar (#10612)
* grammar : handle maxItems == 0 in JSON schema (#13117)
Co-authored-by: Richard Lyons <frob@cloudstaff.com>
* grammar-parser : fix possible null-deref (#9004)
Fixes: https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=70680
Signed-off-by: David Korczynski <david@adalogics.com>
* llama : fix typo in llama-grammar.h [no ci] (#11816)
* * server: fix "--grammar-file" parameter (#12285)
* common : use std::string_view now that we target c++17 (#14319)
* json : support `enum` values within `allOf` (#15830)
* grammar : use int64_t to avoid int overflows in int schema to grammar conversion logic (#16626)
* grammar : support array references in json schema (#16792)
* grammar : support array references in json schema
* Update json-schema-to-grammar.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* grammar : improve regex when naming ref derived rules
* grammar : replace non-conformant definitions array with anyOf test case
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
# Conflicts:
# tests/test-json-schema-to-grammar.cpp
* merge fix
* llama : minor grammar refactor (#10897)
* llama: fix error on bad grammar (#12628)
* grammar : fix integer overflow (#17381)
* Fix DoS / integer overflow
* Remove optional, use INT64_MAX instead as placeholder value (it's technically -1, so it fits :)
* White space
* Actually, since it's unsigned, use UINT64_MAX
# Conflicts:
# src/llama-grammar.cpp
* grammar: fix regression caused by #17381 (#17412)
* grammar: fix regression caused by #17381
* more readable
# Conflicts:
# src/llama-grammar.cpp
* Merge Fix
* Fix warnings
---------
Signed-off-by: David Korczynski <david@adalogics.com>
Co-authored-by: Joe Eli McIlvain <joe.eli.mac@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: frob <rick+github@frob.com.au>
Co-authored-by: Richard Lyons <frob@cloudstaff.com>
Co-authored-by: DavidKorczynski <david@adalogics.com>
Co-authored-by: Daniel Bevenius <daniel.bevenius@gmail.com>
Co-authored-by: firecoperana <firecoperana>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Co-authored-by: Aldehir Rojas <hello@alde.dev>
Co-authored-by: Olivier Chafik <olivier.chafik@gmail.com>
Co-authored-by: Piotr Wilkin (ilintar) <piotr.wilkin@syndatis.com>
Co-authored-by: Xuan-Son Nguyen <son@huggingface.co>
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* model : Granite docling + Idefics3 preprocessing (SmolVLM) (#16206)
* feat: Add granite-docling conversion using trillion pretokenizer
Branch: gabe-l-hart/GraniteDocling
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Add granite-docling vocab pre enum
Branch: gabe-l-hart/GraniteDocling
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Use granite-docling pre
Branch: gabe-l-hart/GraniteDocling
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Add clip_is_idefics3
Branch: gabe-l-hart/GraniteDocling
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Allow multi-token boundary sequences for image templating
Branch: gabe-l-hart/GraniteDocling
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Add tiling support for idefices3 in clip.cpp
This should likely be moved into llava_uhd::get_slice_instructions, but for
now this avoids disrupting the logic there.
Branch: gabe-l-hart/GraniteDocling
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Partial support for full templating for idefics3 in mtmd
There are still errors encoding some of the image chunks, but the token
sequence now matches transformers _almost_ perfectly, except for the double
newline before the global image which shows up as two consecutive newline
tokens instead of a single double-newline token. I think this is happening
because the blocks are tokenized separately then concatenated.
Branch: gabe-l-hart/GraniteDocling
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Fully working image preprocessing for idefics3 w/ resize and slicing
Branch: gabe-l-hart/GraniteDocling
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Parse the preprocessor config's longest side and add it to the mmproj hparams
Branch: GraniteDocling
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Use the longest side instead of size * scale_factor
For Granite Docling, these come out to the same value, but that was just a
conicidence.
Branch: GraniteDocling
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Allow batch encoding and remove clip_is_idefics3
Branch: GraniteDocling
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* refactor: Remove unnecessary conditionals for empty token vectors
Branch: GraniteDocling
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* refactor: Use image_manipulation util
Branch: GraniteDocling
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* add test model
---------
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
# Conflicts:
# convert_hf_to_gguf.py
# convert_hf_to_gguf_update.py
# gguf-py/gguf/constants.py
# gguf-py/gguf/gguf_writer.py
# src/llama-vocab.cpp
# src/llama-vocab.h
* mtmd : support home-cooked Mistral Small Omni (#14928)
* model : add LightOnOCR-1B model (#16764)
* model : add LightOnOCR-1B model
* add test
# Conflicts:
# convert_hf_to_gguf.py
# gguf-py/gguf/constants.py
* mtmd : fix idefics3 preprocessing (#16806)
* mtmd : fix idefics3 preprocessing
* disable granite test
* fix test for granite
* model: Add support for CogVLM model (#15002)
* Added GGUF mappings for CogVLM model
* Add tensor mapping for CogVLM visual encoder
* Add CogVLM to conversion script, no vision part yet
* Added CogVLM vision model to conversion script
* Add graph for CogVLM CLIP model
* Add graph for CogVLM
* Fixes for CogVLM. Now compiles.
* Model now runs
* Fixes for cogvlm graph
* Account for graph context change after rebase
* Changes for whitespace
* Changes in convert script according to comments
* Switch CogVLM LLM graph to merged QKV tensor
* Use rope_type variable instead of direct definition
* Change CogVLM CLIP encoder to use SWIGLU
* Switch CogVLM CLIP to use merged QKV
* Apply rebase edits and remove ggml_cont call that is now unnecessary
* clean up
---------
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
# Conflicts:
# convert_hf_to_gguf.py
# examples/mtmd/clip.cpp
# gguf-py/gguf/constants.py
# gguf-py/gguf/tensor_mapping.py
# src/llama-arch.cpp
# src/llama-arch.h
# src/llama-model.cpp
# src/llama-model.h
* mtmd: refactor preprocessing + support max/min pixels (#16878)
* mtmd: refactor preprocessing + support max/min pixels
* fix mlp type
* implement mix/max pixels
* improve hparams
* better image preproc for qwen
* fix
* fix out of bound composite
* fix (2)
* fix token calculation
* get_merge_kernel_size()
* fix llama4 and lfm2
* gonna fix them all
* use simple resize for qwen
* qwen: increase min tokens
* no resize if dst size == src size
* restore to initial min/max tokens value for qwen
# Conflicts:
# examples/mtmd/clip.cpp
* clip : use FA (#16837)
* clip : use FA
* cont : add warning about unsupported ops
* implement "auto" mode for clip flash attn
* clip : print more detailed op support info during warmup
* cont : remove obsolete comment [no ci]
* improve debugging message
* trailing space
* metal : remove stray return
---------
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
* model: add Janus Pro for image understanding (#16906)
* Add support for Janus Pro
* Update gguf-py/gguf/tensor_mapping.py
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update gguf-py/gguf/tensor_mapping.py
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Address reviewer suggestions
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Add JANUS_PRO constant
* Update clip model handling
Co-authored-by: Xuan-Son Nguyen <son@huggingface.co>
* Update tools/mtmd/clip.cpp
Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com>
* Refactor JANUS_PRO handling in clip.cpp
Co-authored-by: Xuan-Son Nguyen <son@huggingface.co>
* Update tools/mtmd/clip.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* em whitespace
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Co-authored-by: Xuan-Son Nguyen <son@huggingface.co>
Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com>
# Conflicts:
# convert_hf_to_gguf.py
# gguf-py/gguf/constants.py
# gguf-py/gguf/tensor_mapping.py
* mtmd: pad mask for qwen2.5vl (#16954)
* mtmd: pad mask for qwen2.5vl
* improve
* mtmd: add --image-min/max-tokens (#16921)
* mtmd: improve struct initialization (#16981)
* mtmd: allow QwenVL to process larger image by default (#17020)
* Disable flash attention
* mtmd : fix embedding size for image input (#17123)
* mtmd: fix patch_size initialized to random value in audio models (#17128)
* mtmd: fix patch_size initialized to random value in audio models
* add default hparams
* add llama_model_n_embd_inp
* Fix load qwen3 vl
Change batch size
* Add description
* Fix cli build error
---------
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
Co-authored-by: Gabe Goodhart <ghart@us.ibm.com>
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
Co-authored-by: Tianyue-Zhao <zhaotianyue@outlook.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: Zhiyong Wang <85110830+ravenouse@users.noreply.github.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com>
Co-authored-by: firecoperana <firecoperana>
* Fixing Gigachat support
* Gigachat: CUDA FA (needs 192 x 192 for MLA = 3)
* Gigachat: CPU FA (needs 192 x 192 for MLA = 3)
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* Add mainline compatible FA command line option
* Graph reuse: add command line argument to turn it on
* WIP
* This seems to work
* This is perhaps cleaner
* Change the command line option to -gr
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
so more recent users that haven't followed the history of FlashMLA
evolution and hence don't know about the MLA options get the best setting
without having to add -mla 3 on the command line.
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* Fuse concat and copy into K cache
* Avoid ggml_cont() when n_token = 1
Combined effect: about +2% in TG performance with full GPU offload
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* Merge Q and K into a single tensor
* Make V mul mat follow QK mul mat
so they can be fused, which gives a slightly bbetter TG performance.
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* server: add support for vision model
webui: add support for vision model
* server : remove hack for extra parallel slot#10187
* llama : fix KV shift for qwen2vl #13870
* add no-context-shift parameter
---------
Co-authored-by: firecoperana <firecoperana>
* Introducing rope cache
When computing RoPE, the rotation angles in each layer
are exactly the same, and only depend on the token positions
(and other constant, model dependent parameters).
So, I wonder, why don't we compute the angles just once
and then reuse for the Q and K RoPE in each layer?
This commit does it as a POC on the CPU, and uses it in
the Qwen3-MoE compute graph.
* cuda: neox works
* WIP
* rope_cache: norm works
* Fused rope+rope
* Fused rope+rope (norm)
* Fused rms+rms+rope+rope (neox) - not working
* WIP
* Also qwen3
* Add command line arg to disable rope cache
* Disable RoPE cache if rope type is not neox or norm
* Add missing break after merge with main
* Fused fused_rms+fused_rms+rope+rope (with -mqkv)
* Fused fused_rms+fused_rms+rope+rope (without -mqkv)
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* Biased mmvq: minor optimization
* Fusing Q and K rms_norm for TG on CUDA
* Remove commented out code
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* POC: merge Q, K, V into a single, contiguous tensor
Done just for Qwen3-MoE, where I see a 4% uplift in TG.
PP performance gain is sub-percent, if any.
Still, it seems it makes sense to do it in general given
the TG performance gain.
* WIP
* merge_qkv: it works for gpt-oss
...but we see a smaller TG gain (~1.5%)
* WIP
* Don't ignore the return value of create_tensors()
else, when q, k, v get merged and we are running on the CPU,
we get a crash because the backend is trying to use mmap,
but that no longer works.
* merge_qkv: bias can be required, optional, or mandatory
* merge_qkv: glm4.5moe
* merge_qkv: add command loine argument to enable
* merge_qkv: fix tensor dimensions
* merge_qkv: llama-4
* merge_qkv: qwen3 (dense)
* merge_qkv: simplify build_qwen3moe
* cohere2 - simplify graph building
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* Fuse Q, K, V gemv+add
* More gemv+add fusing
* Faster copy when tensors are contiguous
Relevant for storing data into the KV cache. I see ~1% speedup
for fast models (Ling-mini-2.0, gpt-oss-20b, etc.)
* Cleanup
* Make sure the bias really is 1 row to use fusion
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* Change fmoe to be on by default
* Change default fmoe also in llama-bench
* Change flash attention to be on by default
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* Adding fused mul+multi_add + CPU implementation
* fused mul+multi_add: command line argument to disable it
* Faster tensor name formatting
We gain ~1% for Ling-mini-2.0 when running on CUDA.
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* Adding fused mul+multi_add + CPU implementation
* fused mul+multi_add: CUDA
* fused mul+multi_add: command line argument to disable it
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Kawrakow
Samuel Oliveira Alves
joelfarthing
dmaivel
firecoperana
markaalonzo
markaalonzo
Nexes the Elder
dungquixote42
gapeleon
usrlocalben
saood06
firecoperana
Thireus ☠