Qwen3.8-27B-FP8-dynamic

Qwen3.8-27B-FP8-dynamic

FP8 quantization of Qwen/Qwen3.8-27B in the compressed-tensors format:

• Weights: FP8 (E4M3), per-channel scales
• Activations: FP8 (E4M3), per-token dynamic scales
• Kept in BF16: linear_attn (Gated DeltaNet layers), the vision encoder, embed_tokens, lm_head, and the MTP head (mtp.* tensors are carried over unchanged so speculative decoding stays possible)

Only Linear modules inside the transformer blocks are quantized. This is the same recipe RedHatAI uses for its -FP8-dynamic releases, applied with LLM Compressor 0.13.0. No calibration data is needed for this scheme.

Size on disk: ~37 GB. FP8: 18.8 GB (MLP 17.1 GB, full-attention projections 1.7 GB). BF16: 18.0 GB (Gated DeltaNet projections 11.1 GB, embeddings + lm_head 5.1 GB, vision 0.9 GB, MTP 0.85 GB). The BF16 source is 55.6 GB. With TP=2 that is ~18.4 GB of weights per GPU.

This is a community quantization, not an official Qwen release.

Why per-channel FP8 (and not block-128)

The official Qwen/Qwen3.8-27B-FP8 uses block-128 FP8 (weight_block_size [128, 128]). In vLLM, the CUTLASS block-scaled FP8 GEMM requires SM90+ (Hopper) and DeepGEMM requires SM90/SM100, so on Ada (SM 8.9, e.g. RTX 6000 Ada, L40S, RTX 4090) block-128 FP8 falls back to a Triton kernel. Per-channel FP8 with dynamic per-token activations uses the CUTLASS FP8 GEMM on SM 8.9 (CUDA >= 12.4) and also runs on Hopper and Blackwell.

Source: vLLM v0.19.0, csrc/quantization/w8a8/cutlass/scaled_mm_entry.cu (cutlass_scaled_mm_supports_fp8 vs cutlass_scaled_mm_supports_block_fp8).

Deployment with vLLM

Requires vLLM >= 0.17.0.

vllm serve AzatAI/Qwen3.8-27B-FP8-dynamic \
  --tensor-parallel-size 2 \
  --reasoning-parser qwen3 \
  --enable-auto-tool-choice --tool-call-parser qwen3_coder \
  --enable-prefix-caching \
  --speculative-config '{"method": "mtp", "num_speculative_tokens": 3}' \
  --max-num-batched-tokens 32768 \
  --default-chat-template-kwargs '{"enable_thinking": false}'

The MTP head is included (model_mtp.safetensors), so --speculative-config with method: mtp works out of the box; vLLM resolves the drafter as Qwen3_5MTP and shares the target's embeddings and lm_head. Under speculative decoding vLLM ignores min_p and logit_bias.

Thinking is on by default in the chat template; the flag above turns it off by default, and clients re-enable it per request with chat_template_kwargs={"enable_thinking": true} (vLLM >= 0.19 returns it in message.reasoning). Drop the flag to keep the upstream default.

Measured on 2x RTX 6000 Ada (Ada, SM 8.9), vLLM 0.19.0, TP=2

Single stream, greedy unless noted, max_tokens=512:

MTP acceptance: mean 2.56 tokens per step on prose (per-position 0.75 / 0.49 / 0.33), 3.6 on code. Decode without MTP is memory-bandwidth-bound on Ada (~18.4 GB of weights per GPU per token). vLLM selects CutlassFP8ScaledMMLinearKernel for the linear layers on SM 8.9.

Creation

from compressed_tensors.utils import save_mtp_tensors_to_checkpoint
from transformers import AutoProcessor, Qwen3_5ForConditionalGeneration
from llmcompressor import oneshot
from llmcompressor.modifiers.quantization import QuantizationModifier

MODEL_ID = "Qwen/Qwen3.8-27B"
SAVE_DIR = "Qwen3.8-27B-FP8-dynamic"

model = Qwen3_5ForConditionalGeneration.from_pretrained(MODEL_ID, dtype="auto")
processor = AutoProcessor.from_pretrained(MODEL_ID)

recipe = QuantizationModifier(
    targets="Linear",
    scheme="FP8_DYNAMIC",
    ignore=[
        "re:.*lm_head",
        "re:visual.*",
        "re:model.visual.*",
        "re:.*mlp.gate$",
        "re:.*embed_tokens$",
        "re:.*shared_expert_gate$",
        "re:.*linear_attn.*",
    ],
)
oneshot(model=model, recipe=recipe)
model.save_pretrained(SAVE_DIR)
processor.save_pretrained(SAVE_DIR)
save_mtp_tensors_to_checkpoint(MODEL_ID, SAVE_DIR)

Run on CPU (no GPU needed for FP8_DYNAMIC); ~250 GB of RAM headroom is comfortable for the 55.6 GB BF16 source.

Evaluation

No evaluation suite has been run on this checkpoint yet. RedHatAI reports near-identical scores for the same recipe on the Qwen3.6-35B-A3B sibling; treat that as an indication, not a measurement for this model.

License

Apache-2.0, inherited from the base model. The upstream LICENSE file is included unchanged.