* llama-server: recursive GGUF loading
Replace flat directory scan with recursive traversal using
std::filesystem::recursive_directory_iterator. Support for
nested vendor/model layouts (e.g. vendor/model/*.gguf).
Model name now reflects the relative path within --models-dir
instead of just the filename. Aggregate files by parent
directory via std::map before constructing local_model
* server : router config POC (INI-based per-model settings)
* server: address review feedback from @aldehir and @ngxson
PEG parser usage improvements:
- Simplify parser instantiation (remove arena indirection)
- Optimize grammar usage (ws instead of zero_or_more, remove optional wrapping)
- Fix last line without newline bug (+ operator instead of <<)
- Remove redundant end position check
Feature scope:
- Remove auto-reload feature (will be separate PR per @ngxson)
- Keep config.ini auto-creation and template generation
- Preserve per-model customization logic
Co-authored-by: aldehir <aldehir@users.noreply.github.com>
Co-authored-by: ngxson <ngxson@users.noreply.github.com>
* server: adopt aldehir's line-oriented PEG parser
Complete rewrite of INI parser grammar and visitor:
- Use p.chars(), p.negate(), p.any() instead of p.until()
- Support end-of-line comments (key=value # comment)
- Handle EOF without trailing newline correctly
- Strict identifier validation ([a-zA-Z_][a-zA-Z0-9_.-]*)
- Simplified visitor (no pending state, no trim needed)
- Grammar handles whitespace natively via eol rule
Business validation preserved:
- Reject section names starting with LLAMA_ARG_*
- Accept only keys starting with LLAMA_ARG_*
- Require explicit section before key-value pairs
Co-authored-by: aldehir <aldehir@users.noreply.github.com>
* server: fix CLI/env duplication in child processes
Children now receive minimal CLI args (executable, model, port, alias)
instead of inheriting all router args. Global settings pass through
LLAMA_ARG_* environment variables only, eliminating duplicate config
warnings.
Fixes: Router args like -ngl, -fa were passed both via CLI and env,
causing 'will be overwritten' warnings on every child spawn
* add common/preset.cpp
* fix compile
* cont
* allow custom-path models
* add falsey check
* server: fix router model discovery and child process spawning
- Sanitize model names: replace / and \ with _ for display
- Recursive directory scan with relative path storage
- Convert relative paths to absolute when spawning children
- Filter router control args from child processes
- Refresh args after port assignment for correct port value
- Fallback preset lookup for compatibility
- Fix missing argv[0]: store server binary path before base_args parsing
* Revert "server: fix router model discovery and child process spawning"
This reverts commit e3832b42eeea7fcb108995966c7584479f745857.
* clarify about "no-" prefix
* correct render_args() to include binary path
* also remove arg LLAMA_ARG_MODELS_PRESET for child
* add co-author for ini parser code
Co-authored-by: aldehir <hello@alde.dev>
* also set LLAMA_ARG_HOST
* add CHILD_ADDR
* Remove dead code
---------
Co-authored-by: aldehir <aldehir@users.noreply.github.com>
Co-authored-by: ngxson <ngxson@users.noreply.github.com>
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
Co-authored-by: aldehir <hello@alde.dev>
Previously, cmake was forcing `_WIN32_WINNT=0x0A00` for MinGW builds,
This caused "macro redefined" warnings with toolchains that define the version.
This also removes the `GGML_WIN_VER` variable as it is no longer needed.
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
This commit removes the `--backend-dist` option and instead uses the
configured --samplers chain to determine which samplers run on the
backend.
Backend sampling is still enabled using With `--backend_sampling`, and
the sampler chain, either explictly specified using `--samplers` or the
default, is automatically analyzed to determine which samplers can run
on the backend. The system finds the longest contiguous chain of
backend supported samplers from the start of the sampler sequence.
For example:
* If the chain is `top-k -> temperature -> top-p`, and both `top-k` and
`temperature` are backend-supported but `top-p` is not, then `top-k`
and `temperature` will run on the backend, while `top-p` and
subsequent samplers run on the CPU.
* If all configured samplers are supported, the final distribution
sampling will also happen on the backend, transferring only the
sampled token IDs back to the host.
* If the sampler chain starts with an unsupported sampler (e.g.,
`penalties`), all sampling runs on the CPU. Note that this is
currently the case with the default sampler so to use backend sampling
it is required to specify a sampler chain. See below for an example.
The following shows how llama-cli can be run with backend sampling:
```console
$ llama-cli -m models/Qwen2.5-VL-3B-Instruct-Q8_0.gguf \
--prompt 'What is the capital of Sweden?' \
-n 20 \
-no-cnv \
--verbose-prompt \
-ngl 40 \
--backend-sampling \
--samplers 'top_k;temperature'
```
In this case the all sampling will happen on the backend since both
`top_k` and `temperature` are supported backend samplers.
To enable a partial backend sampling (hybrid sampling), for example
running `top_k` and `temperature` on the backend and `typ_p` on the CPU
the following sampler chain could be specified:
```console
$ llama-cli -m models/Qwen2.5-VL-3B-Instruct-Q8_0.gguf \
--prompt 'What is the capital of Sweden?' \
-n 20 \
-no-cnv \
--verbose-prompt \
-ngl 40 \
--backend-sampling \
--samplers 'top_k;temperature;top_p'
```
If this looks good then I'll follow up with updates the llama-cli and
llama-server documentation to reflect these changes.
This commit refactors the model loading process in common/common.cpp
to enable backend sampler to be configure prior to the llama_context
creation.
The motivation for this change is that just being able to set/reset the
backend samplers after the llama_context has been created will cause a
resize to occur in llama_context::output_reserve which we want to avoid.
This commit adds support for performing sampling operations on the
backend (e.g. GPU) as part of the model computation graph.
The motivation for this feature is to enable sampling to be performed
directly on the backend as part of the computation graph being executed,
allowing for some or all of the sampling to be done on the backend.
For example, the backend sampler chain might select/sample a token
directly in which case only the sampled token needs to be transferred
from device memory to host memory.
It is also possible for the backend samplers to perform filtering of
the logits, or compute and filter the probability distribution, in
which case only the filtered logits or probabilites need to be
transferred back to system memory for further processing by CPU
samplers.
Currently the backend sampling works in a similar manner to how
pooling works, it is a function that is called by build_graph and the
sampler operations become part of the models computation graph.
* arg: add --cache-list argument to list cached models
* new manifest naming format
* improve naming
* Update common/arg.cpp
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* minor : code style
* server : fix prompt similarity calculation
* server : initial host-memory prompt caching
* cont
* server : refactor
* cont
* cont : make the server task of the slot const
* cont : minor [no ci]
* server : cache prompts and checkpoints only for completion tasks
* server : improve prompt caching logic
* cont : fix check for number of cached prompts [no ci]
* server : improve caching logic, add -cram CLI arg
* server : print prompt mismatch info
* cont : better naming [no ci]
* server : improve prompt cache loading logic
* server : add option to debug the slot contents (#16482)
* server : add option to debug the slot contents
* Update tools/server/server.cpp
---------
Co-authored-by: Xuan-Son Nguyen <son@huggingface.co>
* server : add option to disable prompt cache
---------
Co-authored-by: Xuan-Son Nguyen <son@huggingface.co>
* 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>
* implement --no-host to disable host buffer
* fix equal_mparams
* move no-host enumeration order together with other model params
---------
Co-authored-by: slaren <slarengh@gmail.com>
* initial commit for branch 3
* generalize `swa_checkpoint` to `ctx_checkpoint`
this extends `llama-server`'s SWA checkpointing logic to include
hybrid/recurrent models such as Jamba, Granite
* oops
* disable debug prints
* keep backwards compat with `--swa-checkpoints`
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* update prompt re-processing message
* fix off-by-one error per GG
* keep `seq_rm` log per GG
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* server : fix checkpoint logic to support recurrent caches
* server : cleanup and fixes
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* add grok-2 support
* type fix
* type fix
* type fix
* "fix" vocab for invalid sequences
* fix expert tensor mapping and spaces in vocab
* add chat template
* fix norm tensor mapping
* rename layer_out_norm to ffn_post_norm
* ensure ffn_post_norm is mapped
* fix experts merging
* remove erroneous FFN_GATE entry
* concatenate split tensors and add more metadata
* process all expert layers and try cat instead of hstack
* add support for community BPE vocab
* fix expert feed forward length and ffn_down concat
* commit this too
* add ffn_up/gate/down, unsure if sequence is right
* add ffn_gate/down/up to tensor names
* correct residual moe (still not working)
* mess--
* fix embedding scale being applied twice
* add built in chat template
* change beta fast for grok if default value
* remove spm vocab in favor of community bpe vocab
* change attention temp length metadata type to integer
* update attention temp length metadata
* remove comment
* replace M_SQRT2 with std::sqrt(2)
* add yarn metadata, move defaults to hparams
To pull and run models via: llama-server -dr gemma3
Add some validators and sanitizers for Docker Model urls and metadata
Signed-off-by: Eric Curtin <eric.curtin@docker.com>
* server : enable /slots by default and make it secure
ggml-ci
* server : fix tests to pass `--no-slots` when necessary
* server : extend /props with info about enabled endpoints
* server : add SWA checkpoints
ggml-ci
* cont : server clean-up
* server : handle state restore fails
* llama : add extended llama_state_seq_ API
* server : do not make checkpoints if --swa-full
ggml-ci
* llama : remove flags value for NONE
* server : configure number of SWA checkpoints with CLI arg
ggml-ci
* args : fix scope of new argument
* examples/finetune -opt SGD (stochastic gradient descent) memory opt
add unit tested GGML_OPT_OPTIMIZER_SGD to ggml - avoids allocating
m, v tensors.
support finetune.cpp arg -opt SGD (or sgd). (default adamw as before)
llama 3.2-1b-F32 result: observed 11gb gpu ram (41 sec/epoch)
when using SGD instead of 19gb (55 sec/epoch) using adamw.
(wikipedia 100 lines finetune)
(
using the same GPU memory, adamw can only do before OOM 512
batch/context, reaching:
train: [███████▉] data=0000140/0000140 loss=0.02575±0.00099 acc=99.52±0.03% t=00:00:47 ETA=00:00:00
val: [███████▉] data=0000008/0000008 loss=4.76565±0.28810 acc=41.46±0.77% t=00:00:00 ETA=00:00:00
SGD is superior, though it converges slower, with max before OOM 1728
batch/context (esp see the better validation perf):
train: [███████▉] data=0000039/0000039 loss=0.00371±0.00010 acc=99.96±0.01% t=00:00:41 ETA=00:00:00
val: [███████▉] data=0000003/0000003 loss=5.11406±0.76034 acc=48.01±0.69% t=00:00:01 ETA=00:00:00
)
note: when finetuning long enough (or w/ enough -lr),
validation accuracy *eventually* drops ('catastrophic forgetting')
-lr-half (halflife) option useful for SGD to avoid oscillation or
super slow underdamped learning (makes setting -lr more forgiving).
terminal -lr for now is set by lr-halvings i.e. if you want at most
1/8 the inital -lr you set -lr-halvings 3.
note: objective loss not directly comparable between adamw, sgd? -
check perplexity or accuracy or consider relative improvements
for convergence
new finetune args -wd 1e-9 to enable weight decay in sgd or adamw,
and max -epochs N (default 2 as before)
cache (1 - wd*alpha) in 'adamw' opt struct -
no noticeable perf benefit, disabled (still done
for new SGD though)
since opt. memory is pre-allocated, the ggml_opt_get_optimizer_params
would probably be able to change between SGD and AdamW with each epoch
but would need to use adamw for the first (unconfirmed - no cmdline arg
to set such a policy yet)
test-opt checks adamw as before and now sgd (except for a few disabled
tests for sgd only; probably just needs logging values and adding
alternate reference values); tolerance on the 'regression'
test is broader for sgd (so we don't need many more epochs)
* Vulkan: Implement GGML_OP_OPT_STEP_SGD
* tests: Fix OPT_STEP_SGD test-backend-ops
* SGD op param store weight-decay and not 1-alpha*wd
* minor + cosmetic changes
* fix vulkan sgd
* try CI fix
---------
Co-authored-by: 0cc4m <picard12@live.de>
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Checkpoint from VS Code for coding agent session
* Initial plan
* Fix typo in --override-tensor-draft flag implementation
* Add null termination for speculative tensor buffer overrides
* Apply suggestions from code review
* Apply suggestions from code review
* Extract tensor override parsing logic to common function (addresses @slaren's feedback)
* Apply suggestions from code review
* Apply suggestions
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: Diego Devesa <slarengh@gmail.com>
Georgi Gerganov
Xuan-Son Nguyen
Pascal
Sigbjørn Skjæret
Daniel Bevenius
Adrien Gallouët
Aaron Teo
Gadflyii
ddh0
jacekpoplawski
Eric Curtin
Johannes Gäßler
Jie Fu (傅杰)
Jonathan Graehl
Copilot
Sachin Desai