Downloads · 30 days
26
15% of all-time downloads
xinjiahui/Qwen3-Coder-30B-A3B-Instruct-FP8-EAGLE3
Qwen3-Coder-30B-A3B-Instruct-FP8-EAGLE3 is a machine learning model from xinjiahui. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
The Eagle3 draft model was trained using the SpecForge framework for the Qwen3-Coder-30B-A3B-Instruct-FP8 model, leveraging a combination of OPC datasets.
Downloads · 30 days
26
15% of all-time downloads
All-time downloads
177
Public
Parameters
141M
848 MB on disk
Likes
1
Public
Click a slice to open those files.
.safetensors283 MB · 100%
How the weights are stored.
BF16141M · 100%
From the Hugging Face model README
The Eagle3 draft model was trained using the SpecForge framework for the Qwen3-Coder-30B-A3B-Instruct-FP8 model, leveraging a combination of OPC datasets.
{
"humaneval": [
{
"batch_size": 1,
"steps": null,
"topk": null,
"num_draft_tokens": null,
"metrics": [
{
"latency": 409.72047470510006,
"output_throughput": 222.746983942377,
"accept_length": 5.334580313303718,
"accuracy": 0.0,
"num_questions": 164,
"num_valid_predictions": 164,
"categorical_performance": null
}
],
"num_samples": null
}
]
}
{
"mtbench": [
{
"batch_size": 1,
"steps": null,
"topk": null,
"num_draft_tokens": null,
"metrics": [
{
"latency": 473.0768761308864,
"output_throughput": 160.5087964159795,
"accept_length": 3.4988941111418304,
"accuracy": null,
"num_questions": 80,
"num_valid_predictions": 0,
"categorical_performance": null
}
],
"num_samples": null
}
]
}
You can use this Eagle3 draft model in SGLang with the following command:
python3 -m sglang.launch_server \
--model <Qwen3-Coder-30B-A3B-Instruct-FP8> \
--speculative-algorithm EAGLE3 \
--speculative-draft-model-path <Qwen3-Coder-30B-A3B-Instruct-FP8-EAGLE3> \
--speculative-num-steps 5 \
--speculative-eagle-topk 8 \
--speculative-num-draft-tokens 32 \
--mem-fraction-static 0.75 \
--tp 8 \
--enable-ep-moe \
--context-length 128000 \
--trust-remote-code \
--host 0.0.0.0 \
--port 30000 \
--dtype bfloat16