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>
* common: move download functions to download.(cpp|h)
* rm unused includes
* minor cleanup
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Add --embd-output-format raw for plain numeric embedding output
This new option outputs embeddings as raw space-separated floats, without JSON or 'embedding N:' prefixes. Useful for downstream vector pipelines and scripting.
* Move raw output handling into format handling section
* Move raw output handling into else-if block with other format handlers
* Use LOG instead of printf for raw embedding output
* docs: document 'raw' embedding output format in arg.cpp and README
* Update the docs on -t --threads
* Revert "Update the docs on -t --threads"
This reverts commit eba97345e2.
* docs: clarify -t/--threads parameter uses CPU threads and defaults to all available cores
* Update arg.cpp
* 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>
* rpc : add support for multiple devices
Allow rpc-server to expose multiple devices from a single endpoint.
Change RPC protocol to include device identifier where needed.
closes: #15210
* fixes
* use ggml_backend_reg_t
* address review comments
* fix llama-bench backend report
* address review comments, change device naming
* fix cmd order
* 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>
* common: introduce http.h for httplib-based client
This change moves cpp-httplib based URL parsing and client setup into
a new header `common/http.h`, and integrates it in `arg.cpp` and `run.cpp`.
It is an iteration towards removing libcurl, while intentionally
minimizing changes to existing code to guarantee the same behavior when
`LLAMA_CURL` is used.
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
* tools : add missing WIN32_LEAN_AND_MEAN
Signed-off-by: Adrien Gallouët <adrien@gallouet.fr>
---------
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
Signed-off-by: Adrien Gallouët <adrien@gallouet.fr>
`test-arg-parser.cpp` has been updated to work consistently,
regardless of whether CURL or SSL support is available, and
now always points to `ggml.ai`.
The previous timeout test has been removed, but it can be
added back by providing a dedicated URL under `ggml.ai`.
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
The JSON parser is temporarily kept only for backward compatibility. It
reads the etag from old .json files to prevent unnecessary re-downloads
for existing users.
This legacy code can be removed in a future version.
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
* vendor : update httplib
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
* common : use cpp-httplib as a cURL alternative for downloads
The existing cURL implementation is intentionally left untouched to
prevent any regressions and to allow for safe, side-by-side testing by
toggling the `LLAMA_CURL` CMake option.
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
* ggml : Bump to Windows 10
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
---------
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
* common : use the json parser
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
* common : enable --offline mode without CURL support
This change refactors the download logic to properly support offline mode
even when the project is built without CURL.
Without this commit, using `--offline` would give the following error:
error: built without CURL, cannot download model from the internet
even if all the files are already cached.
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
---------
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
These two local variables 'arg' and 'arg_prefix' have been overriden by:
1. for (const auto & arg : opt.args)
2. for (int i = 1; i < argc; i++) {
const std::string arg_prefix = "--";
std::string arg = argv[i];
- Implement resumable downloads in common_download_file_single function
- Add detection of partial download files (.downloadInProgress)
- Check server support for HTTP Range requests via Accept-Ranges header
- Implement HTTP Range request with "bytes=<start>-" header
- Open files in append mode when resuming vs create mode for new downloads
Signed-off-by: Eric Curtin <eric.curtin@docker.com>
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
This commit updates the bash completion script to include the -m
short option for the --model argument.
The motivation for this is that currently tab completion only works the
full --model option, and it is nice to have it work for the short option
as well.
* 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>
Daniel Bevenius
Aaron Teo
Georgi Gerganov
Xuan-Son Nguyen
Sigbjørn Skjæret
Shagun Bera
Sam Malayek
takasurazeem
Pascal
Gadflyii
Radoslav Gerganov
ddh0
Adrien Gallouët
Adrien Gallouët
Haiyue Wang
Eric Curtin
jacekpoplawski
Aman Gupta
Diego Devesa
Eric Curtin
Eric Curtin
Johannes Gäßler
Jie Fu (傅杰)
Gian-Carlo Pascutto
Jonathan Graehl
Copilot