* 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
Purpose:
Add --minilog flag to llama-sweep-bench that filters log output to show only essential GPU/layer distribution information while suppressing verbose model metadata and per-layer device assignment messages.
Changes:
- Add llama_selective_log_callback with blacklist approach (sweep-bench.cpp)
Blacklisted patterns (hidden):
- Per-layer device assignments ('Setting default device in layer')
- KV metadata dump header and entries
- Tensor type counts
- Model validation messages
- EOG/special token cache info
- Metadata printout (llm_load_print_meta, print_info)
- Layer sizes table
- Tensor loading info (llm_load_tensors)
- Separator lines
- Most common cases of incomplete/continuation lines are also hidden
All other log output is shown, including:
- GPU VRAM info
- Split/buffer distribution per device
- Graph split estimates
- Final benchmark table and timings
* 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>
Two changes:
1. Add four missing environment variable bindings to
gpt_params_parse_from_env():
- LLAMA_ARG_CACHE_TYPE_K (string, e.g. "q8_0")
- LLAMA_ARG_CACHE_TYPE_V (string, e.g. "q8_0")
- LLAMA_ARG_MLOCK (bool, "1"/"true")
- LLAMA_ARG_K_CACHE_HADAMARD (bool, "1"/"true")
2. Call gpt_params_parse_from_env() from gpt_params_parse() so that
ALL tools (llama-cli, llama-bench, etc.) respect env vars, not
just llama-server. Env vars act as defaults; CLI flags override.
Follows the existing get_env() pattern and uses the same
LLAMA_ARG_ prefix convention as the other env vars.
Co-authored-by: Pipboyguy <>
---------
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()
* 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
* Fix tool call for Qwen3.5
Loosely based on mainline changes from:
* https://github.com/ggml-org/llama.cpp/pull/19635
* https://github.com/ggml-org/llama.cpp/pull/19765
Also need to change the grammar to allow the model to make multiple
tool calls in a row. This was likely broken for Qwen3 Coder prior to
this commit.
* Fix the grammar for the subsequent parameters after the first one
* 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
* raw parameters.md
* fix small typos in common.cpp
* Update build args in parameters.md
* Update parameters.md
- format as table
- sections
* Update README.md
- quickstart
- build and run
* Update parameters.md
other tools examples
* add PR links
* multiple updates to parameters.md
- description
- add jargon section
- add suggestions from feedbacks
* don't imply that only linux is supported in README.md
* add alias to parameters.md
* Update README.md with recent models and features
* Update parameters.md with latest features
* address suggestions
- no-ooae
- placeholder for common commands
- no-kv-offload
- llama-sweep-bench
- placeholder for unique parameters
* specify Linux distro in README.md
* 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
* 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>
* 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
* 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
* 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
* server: improve speed of speculative decoding
change logs
rpc: add recompute
spec dec fix
* Fix n_batch_size not set to context size for draft model
---------
Co-authored-by: firecoperana <firecoperana>
* 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>
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>
* 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>
* 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>
* Add command line argument for draft model
* Remove second context of draft model
* Format print
* print usage if parsing -draft fails
---------
Co-authored-by: firecoperana <firecoperana>
* server: fix crash when prompt has image and is too long
* server: fix CORS
* server: fix empty result for embedding
* change error message to truncate prompt
* server: fix slot id for save and load state
* bug fix
* server: update slot similarity to handle mtmd
* server: quick hack to calculate number of token processed with image
* server: fix out of range error when detokenizing prompt under verbose
* Add back Access-Control-Allow-Origin
* Server: Add prompt tokens in embedding results
---------
Co-authored-by: firecoperana <firecoperana>
* 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>
* 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>
server : handle models with missing EOS token (#8997)
server : fix segfault on long system prompt (#8987)
* server : fix segfault on long system prompt
* server : fix parallel generation with very small batch sizes
* server : fix typo in comment
server : init stop and error fields of the result struct (#9026)
server : fix duplicated n_predict key in the generation_settings (#8994)
server : support reading arguments from environment variables (#9105)
* server : support reading arguments from environment variables
* add -fa and -dt
* readme : specify non-arg env var
server : add some missing env variables (#9116)
* server : add some missing env variables
* add LLAMA_ARG_HOST to server dockerfile
* also add LLAMA_ARG_CONT_BATCHING
Credits are to the respective authors.
Not a single merge conflict occurred.
Compiled, then tested without bug.
* Add new webui from llama.cpp
* Add new webui
* feat: Improve mobile UI for Settings Dialog (#16084)
* feat: Improve mobile UI for Settings Dialog
* chore: update webui build output
* fix: Linting errors
* chore: update webui build output
# Conflicts:
# examples/server/webui_llamacpp/src/lib/components/app/chat/ChatSettings/ChatSettingsFields.svelte
# examples/server/webui_llamacpp/src/lib/components/app/chat/ChatSettings/ChatSettingsSection.svelte
# tools/server/public/index.html.gz
* webui : fix handling incomplete chunks (#16107)
* Always show message actions for mobile UI + improvements for user message sizing (#16076)
# Conflicts:
# .gitignore
# examples/server/webui_llamacpp/package.json
# examples/server/webui_llamacpp/scripts/dev.sh
# tools/server/webui/scripts/post-build.sh
* webui: switch to hash-based routing (alternative of #16079) (#16157)
* Switched web UI to hash-based routing
* Added hash to missed goto function call
* Removed outdated SPA handling code
* Fixed broken sidebar home link
# Conflicts:
# examples/server/webui_llamacpp/src/routes/+layout.ts
# tools/server/server.cpp
* Allow viewing conversations even when llama server is down (#16255)
* webui: allow viewing conversations and sending messages even if llama-server is down
- Cached llama.cpp server properties in browser localStorage on startup, persisting successful fetches and reloading them when refresh attempts fail so the chat UI continues to render while the backend is unavailable.
- Cleared the stored server properties when resetting the store to prevent stale capability data after cache-backed operation.
- Kept the original error-splash behavior when no cached props exist so fresh installs still surface a clear failure state instead of rendering stale data.
* feat: Add UI for `props` endpoint unavailable + cleanup logic
* webui: extend cached props fallback to offline errors
Treat connection failures (refused, DNS, timeout, fetch) the same way as
server 5xx so the warning banner shows up when cache is available, instead
of falling back to a full error screen.
* webui: Left the chat form enabled when a server warning is present so operators can keep sending messages
e.g., to restart the backend over llama-swap, even while cached /props data is in use
* chore: update webui build output
---------
Co-authored-by: Pascal <admin@serveurperso.com>
# Conflicts:
# examples/server/webui_llamacpp/src/lib/components/app/chat/ChatScreen/ChatScreenWarning.svelte
# examples/server/webui_llamacpp/src/lib/constants/localstorage-keys.ts
* Enhance text file detection logic for file attachments (#16199)
* feat: Enhances text file detection logic
* chore: Build static `webui` output
* chore: update webui build output
# Conflicts:
# examples/server/webui_llamacpp/src/lib/constants/binary-detection.ts
* Show message actions by default (#16289)
* fix: preserved zero values in chat settings inputs and textareas by switching to nullish coalescing for field values and default placeholders (#16312)
* Improve Mobile UI for dialogs and action dropdowns (#16222)
* fix: Always show conversation item actions
* feat: Improve Alert Dialog and Dialog mobile UI
* feat: Add settings reset to default confirmation
* fix: Close Edit dialog on save
* chore: update webui build output
* webui: implement proper z-index system and scroll management
- Add CSS variable for centralized z-index control
- Fix dropdown positioning with Settings dialog conflicts
- Prevent external scroll interference with proper event handling
- Clean up hardcoded z-index values for maintainable architecture
* webui: ensured the settings dialog enforces dynamic viewport height on mobile while retaining existing desktop sizing overrides
* feat: Use `dvh` instead of computed px height for dialogs max height on mobile
* chore: update webui build output
* feat: Improve Settings fields UI
* chore: update webui build output
* chore: update webui build output
---------
Co-authored-by: Pascal <admin@serveurperso.com>
* Fix thinking blocks with quotes + add handling `[THINK]...[/THINK]` blocks (#16326)
* fix: prevent reasoning blocks with quotes from being truncated
* chore: update webui build output
* feat: Improve thinking content parsing
* test: Adds ChatMessage component stories for different thinking blocks
* chore: update webui build output
* fix: ChatMessage story fix
---------
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* Chatapi ignore empty sampling (#16330)
* fix: skip empty sampling fields instead of coercing to 0 in chat API options
* chore: update webui build output
* webui: Remove running `llama-server` within WebUI `dev.sh` script (#16363)
* Add optional setting for showing "Model used:" information (#16337)
* feat: Add a setting to include model name used to generate the message
* feat: UI improvements
* feat: Save model info along with the database message entry creation
* chore: Build webui static output
* Improve code block color theming (#16325)
* feat: Improve code block theming
* chore: update webui build output
* chore: Update webui static build
* Conversation action dialogs as singletons from Chat Sidebar + apply conditional rendering for Actions Dropdown for Chat Conversation Items (#16369)
* fix: Render Conversation action dialogs as singletons from Chat Sidebar level
* chore: update webui build output
* fix: Render Actions Dropdown conditionally only when user hovers conversation item + remove unused markup
* chore: Update webui static build
* fix: Always truncate conversation names
* chore: Update webui static build
* fix: track viewportHeight via window.innerHeight to avoid unwanted scrolling (#16356)
Use <svelte:window bind:innerHeight> instead of manual resize listener
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* webui : Fix messages payload sent to chat completions (#16402)
* fix: Include just the currently active message branches instead of all in chat completions request
* chore: Build webui static output
* chore: Formatting
* chore: update webui build output
* Capture model name only after first token (streaming) or completed request (#16405)
* feat: Capture model name only after first token (streaming) or completed request (non-streaming)
* chore: update webui build output
* chore: update webui build output
* Fix missing messages on sibling navigation (#16408)
* fix: resolve message disappearing issue when navigating between regenerated siblings by using current leaf nodes instead of cached sibling IDs
* chore: update webui build output
* chore: update webui build output
* webui : added download action (#13552) (#16282)
* webui : added download action (#13552)
* webui : import and export (for all conversations)
* webui : fixed download-format, import of one conversation
* webui : add ExportedConversations type for chat import/export
* feat: Update naming & order
* chore: Linting
* webui : Updated static build output
---------
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* refactor: centralize CoT parsing in backend for streaming mode (#16394)
* refactor: unify reasoning handling via backend reasoning_content, drop frontend tag parsing
- Updated the chat message component to surface backend-supplied reasoning via message.thinking while showing the raw assistant content without inline tag scrubbing
- Simplified chat streaming to append content chunks directly, stream reasoning into the message model, and persist any partial reasoning when generation stops
- Refactored the chat service SSE handler to rely on server-provided reasoning_content, removing legacy <think> parsing logic
- Refreshed Storybook data and streaming flows to populate the thinking field explicitly for static and streaming assistant messages
* refactor: implement streaming-aware universal reasoning parser
Remove the streaming mode limitation from --reasoning-format by refactoring
try_parse_reasoning() to handle incremental parsing of <think> tags across
all formats.
- Rework try_parse_reasoning() to track whitespace, partial tags, and
multiple reasoning segments, allowing proper separation of reasoning_content
and content in streaming mode
- Parse reasoning tags before tool call handling in content-only and Llama 3.x
formats to ensure inline <think> blocks are captured correctly
- Change default reasoning_format from 'auto' to 'deepseek' for consistent
behavior
- Add 'deepseek-legacy' option to preserve old inline behavior when needed
- Update CLI help and documentation to reflect streaming support
- Add parser tests for inline <think>...</think> segments
The parser now continues processing content after </think> closes instead of
stopping, enabling proper message.reasoning_content and message.content
separation in both streaming and non-streaming modes.
Fixes the issue where streaming responses would dump everything (including
post-thinking content) into reasoning_content while leaving content empty.
* refactor: address review feedback from allozaur
- Passed the assistant message content directly to ChatMessageAssistant to drop the redundant derived state in the chat message component
- Simplified chat streaming updates by removing unused partial-thinking handling and persisting partial responses straight from currentResponse
- Refreshed the ChatMessage stories to cover standard and reasoning scenarios without the old THINK-tag parsing examples
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* refactor: restore forced reasoning prefix to pass test-chat ([chat] All tests passed)
- store the exact sequence seen on input when 'thinking_forced_open' enforces a reasoning block
- inject this prefix before the first accumulated segment in 'reasoning_content', then clear it to avoid duplication
- repeat the capture on every new 'start_think' detection to properly handle partial/streaming flows
* refactor: address review feedback from ngxson
* debug: say goodbye to curl -N, hello one-click raw stream
- adds a new checkbox in the WebUI to display raw LLM output without backend parsing or frontend Markdown rendering
* Update tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessage.svelte
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* webui: add Storybook example for raw LLM output and scope reasoning format toggle per story
- Added a Storybook example that showcases the chat message component in raw LLM output mode with the provided trace sample
- Updated every ChatMessage story to toggle the disableReasoningFormat setting so the raw-output rendering remains scoped to its own example
* npm run format
* chat-parser: address review feedback from ngxson
Co-authored-by: Xuan Son Nguyen <thichthat@gmail.com>
---------
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
Co-authored-by: Xuan Son Nguyen <thichthat@gmail.com>
# Conflicts:
# common/arg.cpp
# examples/server/webui_llamacpp/src/lib/utils/thinking.ts
# tools/server/README.md
* No markdown in cot (#16483)
* fix: let the model think in plaintext
* chore: npm run format + npm run build
* webui: updated the chat service to only include max_tokens in the req… (#16489)
* webui: updated the chat service to only include max_tokens in the request payload when the setting is explicitly provided, while still mapping explicit zero or null values to the infinite-token sentinel
* chore: update webui build output
* feat: render user content as markdown option (#16358)
* feat: render user content as markdown option
- Add a persisted 'renderUserContentAsMarkdown' preference to the settings defaults and info metadata so the choice survives reloads like other options
- Surface the new 'Render user content as Markdown' checkbox in the General section of the chat settings dialog, beneath the PDF toggle
- Render user chat messages with 'MarkdownContent' when the new setting is enabled, matching assistant formatting while preserving the existing card styling otherwise
- chore: update webui build output
* chore: update webui build output
* webui: remove client-side context pre-check and rely on backend for limits (#16506)
* fix: make SSE client robust to premature [DONE] in agentic proxy chains
* webui: remove client-side context pre-check and rely on backend for limits
Removed the client-side context window pre-check and now simply sends messages
while keeping the dialog imports limited to core components, eliminating the
maximum context alert path
Simplified streaming and non-streaming chat error handling to surface a generic
'No response received from server' error whenever the backend returns no content
Removed the obsolete maxContextError plumbing from the chat store so state
management now focuses on the core message flow without special context-limit cases
* webui: cosmetic rename of error messages
* Update tools/server/webui/src/lib/stores/chat.svelte.ts
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* Update tools/server/webui/src/lib/stores/chat.svelte.ts
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* Update tools/server/webui/src/lib/components/app/chat/ChatScreen/ChatScreen.svelte
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* Update tools/server/webui/src/lib/components/app/chat/ChatScreen/ChatScreen.svelte
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* chore: update webui build output
---------
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
# Conflicts:
# examples/server/webui_llamacpp/src/lib/components/app/dialogs/ChatErrorDialog.svelte
# examples/server/webui_llamacpp/src/lib/components/app/dialogs/MaximumContextAlertDialog.svelte
# examples/server/webui_llamacpp/src/lib/services/context.ts
* fix: add remark plugin to render raw HTML as literal text (#16505)
* fix: add remark plugin to render raw HTML as literal text
Implemented a missing MDAST stage to neutralize raw HTML like major LLM WebUIs
do ensuring consistent and safe Markdown rendering
Introduced 'remarkLiteralHtml', a plugin that converts raw HTML nodes in the
Markdown AST into plain-text equivalents while preserving indentation and
line breaks. This ensures consistent rendering and prevents unintended HTML
execution, without altering valid Markdown structure
Kept 'remarkRehype' in the pipeline since it performs the required conversion
from MDAST to HAST for KaTeX, syntax highlighting, and HTML serialization
Refined the link-enhancement logic to skip unnecessary DOM rewrites,
fixing a subtle bug where extra paragraphs were injected after the first
line due to full innerHTML reconstruction, and ensuring links open in new
tabs only when required
Final pipeline: remarkGfm -> remarkMath -> remarkBreaks -> remarkLiteralHtml
-> remarkRehype -> rehypeKatex -> rehypeHighlight -> rehypeStringify
* fix: address review feedback from allozaur
* chore: update webui build output
# Conflicts:
# examples/server/webui_llamacpp/src/lib/constants/literal-html.ts
* Add server-driven parameter defaults and syncing (#16515)
# Conflicts:
# examples/server/webui_llamacpp/src/lib/components/app/chat/ChatSettings/ParameterSourceIndicator.svelte
# examples/server/webui_llamacpp/src/lib/constants/precision.ts
# examples/server/webui_llamacpp/src/lib/services/parameter-sync.spec.ts
# examples/server/webui_llamacpp/src/lib/services/parameter-sync.ts
# examples/server/webui_llamacpp/src/lib/utils/config-helpers.ts
# examples/server/webui_llamacpp/src/lib/utils/precision.ts
* fix: added a normalization step for MathJax-style \[\] and \(\) delimiters (#16599)
* fix: added a normalization step for MathJax-style \[\] and \(\) delimiters
So inline and block equations are converted before KaTeX rendering,
enabling proper display of model-generated LaTeX in the WebUI
* chore: update webui build output
* webui: reorganize settings layout (#16607)
* webui: reorganize settings layout
* chore: update webui build output
* fix: remove unused variable
* chore: update webui build output
* Enable per-conversation loading states to allow having parallel conversations (#16327)
* feat: Per-conversation loading states and tracking streaming stats
* chore: update webui build output
* refactor: Chat state management
Consolidates loading state management by using a global `isLoading` store synchronized with individual conversation states.
This change ensures proper reactivity and avoids potential race conditions when updating the UI based on the loading status of different conversations. It also improves the accuracy of statistics displayed.
Additionally, slots service methods are updated to use conversation IDs for per-conversation state management, avoiding global state pollution.
* feat: Adds loading indicator to conversation items
* chore: update webui build output
* fix: Fix aborting chat streaming
Improves the chat stream abortion process by ensuring that partial responses are saved before the abort signal is sent.
This avoids a race condition where the onError callback could clear the streaming state before the partial response is saved. Additionally, the stream reading loop and callbacks are now checked for abort signals to prevent further processing after abortion.
* refactor: Remove redundant comments
* chore: build webui static output
* refactor: Cleanup
* chore: update webui build output
* chore: update webui build output
* fix: Conversation loading indicator for regenerating messages
* chore: update webui static build
* feat: Improve configuration
* feat: Install `http-server` as dev dependency to not need to rely on `npx` in CI
* Import/Export UX improvements (#16619)
* webui : added download action (#13552)
* webui : import and export (for all conversations)
* webui : fixed download-format, import of one conversation
* webui : add ExportedConversations type for chat import/export
* feat: Update naming & order
* chore: Linting
* feat: Import/Export UX improvements
* chore: update webui build output
* feat: Update UI placement of Import/Export tab in Chat Settings Dialog
* refactor: Cleanup
chore: update webui build output
* feat: Enable shift-click multiple conversation items selection
* chore: update webui static build
* chore: update webui static build
---------
Co-authored-by: Sascha Rogmann <github@rogmann.org>
# Conflicts:
# examples/server/webui_llamacpp/src/lib/components/app/chat/ChatSettings/ConversationSelectionDialog.svelte
# examples/server/webui_llamacpp/src/lib/components/app/chat/ChatSettings/ImportExportTab.svelte
# examples/server/webui_llamacpp/src/lib/utils/conversation-utils.ts
* Prevent premature submission on IME input (#16673)
* fix: Prevent premature submission on IME input
* chore: update webui static build
* refactor: Put IME completion checker in a helper function and add checking for `KeyboardEvent.eventKey === 229`
* chore: update webui static build
* chore: update webui static build
* chore: update webui static build
# Conflicts:
# examples/server/webui_llamacpp/src/lib/utils/is-ime-composing.ts
* Handle legacy 'context' attachments (#16687)
* webui: introduce OpenAI-compatible model selector in JSON payload (#16562)
* webui: introduce OpenAI-compatible model selector in JSON payload
* webui: restore OpenAI-Compatible model source of truth and unify metadata capture
This change re-establishes a single, reliable source of truth for the active model:
fully aligned with the OpenAI-Compat API behavior
It introduces a unified metadata flow that captures the model field from both
streaming and non-streaming responses, wiring a new onModel callback through ChatService
The model name is now resolved directly from the API payload rather than relying on
server /props or UI assumptions
ChatStore records and persists the resolved model for each assistant message during
streaming, ensuring consistency across the UI and database
Type definitions for API and settings were also extended to include model metadata
and the onModel callback, completing the alignment with OpenAI-Compat semantics
* webui: address review feedback from allozaur
* webui: move model selector into ChatForm (idea by @allozaur)
* webui: make model selector more subtle and integrated into ChatForm
* webui: replaced the Flowbite selector with a native Svelte dropdown
* webui: add developer setting to toggle the chat model selector
* webui: address review feedback from allozaur
Normalized streamed model names during chat updates
by trimming input and removing directory components before saving
or persisting them, so the conversation UI shows only the filename
Forced model names within the chat form selector dropdown to render as
a single-line, truncated entry with a tooltip revealing the full name
* webui: toggle displayed model source for legacy vs OpenAI-Compat modes
When the selector is disabled, it falls back to the active server model name from /props
When the model selector is enabled, the displayed model comes from the message metadata
(the one explicitly selected and sent in the request)
* Update tools/server/webui/src/lib/components/app/chat/ChatForm/ChatFormActions.svelte
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* Update tools/server/webui/src/lib/constants/localstorage-keys.ts
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* Update tools/server/webui/src/lib/components/app/chat/ChatForm/ChatFormModelSelector.svelte
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* Update tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessageAssistant.svelte
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* Update tools/server/webui/src/lib/services/chat.ts
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* Update tools/server/webui/src/lib/services/chat.ts
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* webui: refactor model selector and persistence helpers
- Replace inline portal and event listeners with proper Svelte bindings
- Introduce 'persisted' store helper for localStorage sync without runes
- Extract 'normalizeModelName' utils + Vitest coverage
- Simplify ChatFormModelSelector structure and cleanup logic
Replaced the persisted store helper's use of '$state/$effect' runes with
a plain TS implementation to prevent orphaned effect runtime errors
outside component context
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* webui: document normalizeModelName usage with inline examples
* Update tools/server/webui/src/lib/components/app/chat/ChatForm/ChatFormModelSelector.svelte
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* Update tools/server/webui/src/lib/stores/models.svelte.ts
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* Update tools/server/webui/src/lib/stores/models.svelte.ts
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* webui: extract ModelOption type into dedicated models.d.ts
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* webui: refine ChatMessageAssistant displayedModel source logic
* webui: stabilize dropdown, simplify model extraction, and init assistant model field
* chore: update webui static build
* Update tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessageAssistant.svelte
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* chore: npm format, update webui static build
* webui: align sidebar trigger position, remove z-index glitch
* chore: update webui build output
---------
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
# Conflicts:
# examples/server/webui_llamacpp/src/lib/components/app/chat/ChatForm/ChatFormModelSelector.svelte
# examples/server/webui_llamacpp/src/lib/services/models.ts
# examples/server/webui_llamacpp/src/lib/stores/models.svelte.ts
# examples/server/webui_llamacpp/src/lib/stores/persisted.svelte.ts
# examples/server/webui_llamacpp/src/lib/types/models.d.ts
# examples/server/webui_llamacpp/src/lib/utils/model-names.test.ts
# examples/server/webui_llamacpp/src/lib/utils/model-names.ts
# examples/server/webui_llamacpp/src/lib/utils/portal-to-body.ts
* webui: support q URL parameter (#16728)
* webui: support q URL parameter
Fixes#16722
I’ve checked that it works with Firefox’s AI tools
* webui: apply suggestions from code review
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* chore: update webui static build
---------
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
* build fix
---------
Co-authored-by: firecoperana <firecoperana>
Co-authored-by: Aleksander Grygier <aleksander.grygier@gmail.com>
Co-authored-by: Quentin Bramas <quentin.bramas@gmail.com>
Co-authored-by: Isaac McFadyen <isaac@imcf.me>
Co-authored-by: Pascal <admin@serveurperso.com>
Co-authored-by: Sascha Rogmann <59577610+srogmann@users.noreply.github.com>
Co-authored-by: Xuan Son Nguyen <thichthat@gmail.com>
Co-authored-by: Sascha Rogmann <github@rogmann.org>
Co-authored-by: Florian Badie <florianbadie@odrling.xyz>
* 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: CUDA
* fused mul+multi_add: command line argument to disable it
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
* Better argsort (CPU)
* Attemt at grouped topk
* This seems to do the trick for grouped experts routing
* Cleanup
* Trying to merge, something is not right
* Working merged grouped top_k (CPU)
* Add command line option to enable grouped expert routing
* Add grouped expert routing option to llama-bench
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Kawrakow
Samuel Oliveira Alves
firecoperana
Nexes the Elder
dungquixote42
Marcel Coetzee
Yap Sok Ann
TheAIGuyFromAR
mcm007
hksdpc255
firecoperana