Skip to content

[gpt-oss-120b] Add --enable_speculative_decode flag for Interactive scenario #2582

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
viraatc wants to merge 1 commit into
base: master
Choose a base branch
from
Open

[gpt-oss-120b] Add --enable_speculative_decode flag for Interactive scenario #2582

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 12 additions & 0 deletions language/gpt-oss-120b/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -55,6 +55,18 @@ Use [`./sglang/run_server.sh`](./sglang/run_server.sh) to launch an SGLang serve

The script uses `python3 -m sglang.launch_server` tp instantiate the model, with `tp=pp=ep=1`, and `dp` as specified.

#### Speculative Decoding (Interactive scenario)

Pass `--enable_speculative_decode` to enable EAGLE3 speculative decoding for
the Interactive scenario:

```bash
./run_server.sh --enable_speculative_decode
```

Per the Interactive scenario policy, `num_steps=3` and `topk=1` are fixed; the
draft head is [`nvidia/gpt-oss-120b-Eagle3-long-context`](https://huggingface.co/nvidia/gpt-oss-120b-Eagle3-long-context).

You may also use docker:

```bash
Expand Down
22 changes: 18 additions & 4 deletions language/gpt-oss-120b/sglang/run_server.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,6 +8,13 @@ eagle_path=""
stream_interval=500
extra_args=""

# Interactive scenario: EAGLE3 speculative decoding with the long-context head.
# num_steps=3 and topk=1 are fixed per the Interactive scenario policy.
enable_speculative_decode=false
SPECULATIVE_DRAFT_MODEL_PATH="nvidia/gpt-oss-120b-Eagle3-long-context"
SPECULATIVE_NUM_STEPS=3
SPECULATIVE_TOPK=1

while [[ $# -gt 0 ]]; do
case $1 in
--dp)
Expand All @@ -26,6 +33,10 @@ while [[ $# -gt 0 ]]; do
stream_interval=$2
shift 2
;;
--enable_speculative_decode)
enable_speculative_decode=true
shift 1
;;
*)
extra_args="$extra_args $1"
shift 1
Expand All @@ -43,14 +54,17 @@ args=" --model-path $model_path \
--enable-metrics \
--stream-interval $stream_interval "

# Explicit --eagle_path wins over the --enable_speculative_decode defaults so
# existing callers keep working unchanged.
if [ -n "$eagle_path" ]; then
args="$args --speculative-draft-model-path $eagle_path \
--speculative-algorithm EAGLE3"
elif [ "$enable_speculative_decode" = "true" ]; then
args="$args --speculative-draft-model-path $SPECULATIVE_DRAFT_MODEL_PATH \
--speculative-algorithm EAGLE3 \
--speculative-num-steps $SPECULATIVE_NUM_STEPS \
--speculative-eagle-topk $SPECULATIVE_TOPK"
fi

# --speculative-num-steps 1 \
# --speculative-eagle-topk 1 \
# --speculative-num-draft-tokens 3 \

set -x
python3 -m sglang.launch_server "$args" "$extra_args"
Loading