Qwen3.8-27B-FP8

Qwen3.8-27B FP8

This repository is a quantized deployment variant of the official Qwen/Qwen3.8-27B model. It is not a fine-tune.

Provenance and configuration

• Exact parent revision: 1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0
• Quantizer: GPTQModel FP8
• Quantizer revision: 7df3d1e837cb5552512bf99b8729c334d0b2c010
• Quantization configuration: {"bits": 8, "calibration_batch_size": null, "calibration_corpus_sha256": null, "calibration_samples": 0, "calibration_sequence_length": null, "calibration_sort": null, "checkpoint_format": "float8_e4m3fn", "desc_act": false, "format": "float8_e4m3fn", "group_size": -1, "lm_head": false, "meta": {"auto_forward_data_parallel": true, "dense_vram_strategy": "exclusive", "dense_vram_strategy_devices": null, "fallback": {"smooth": null, "strategy": "rtn", "threshold": "0.5%"}, "gc_mode": "interval", "moe_vram_strategy": "exclusive", "moe_vram_strategy_devices": null, "offload_to_disk": false, "offload_to_disk_path": null, "pack_impl": "cpu", "wait_for_submodule_finalizers": false}, "method": "fp8", "pack_dtype": "int32", "quant_method": "fp8", "weight_scale_method": "row", "weight_scale_semantics": "inverse"}
• Artifact size: 29.127 GiB
• Supported runtime: GPTQModel FP8 / Ada scaled GEMM
• Tested hardware: NVIDIA GeForce RTX 4070 12 GB (Ada, SM89), WSL2 Ubuntu 24.04

Calibration and architecture handling

The calibration corpus is original CC0-1.0 material authored for this project. It contains 216 text/tool samples and 8 vision samples across prose, code, debugging, math, science, structured output, tools, multilingual, and long-context categories. This build used None.

• Tensor exclusions: ["*.input_layernorm", "*.post_attention_layernorm", "*.self_attn.q_norm", "*.self_attn.k_norm", "*.linear_attn.norm", "*.linear_attn.conv1d", "*.linear_attn.in_proj_a", "*.linear_attn.in_proj_b"]
• Vision: True (deterministic image generation gate)
• Tool use: True (schema-valid XML tool-call gate)
• MTP: True for preservation of the 15 external tensors; no local speculative drafter/acceptance runtime is claimed.
• Processor/chat template: exact pinned-parent copy verified.

Numerical validation

Full-vocabulary held-out comparison used 64 token positions.

Performance smoke

• Execution mode: RTX 4070 + CPU offload/auto map
• Device map summary: {"cpu": 59, "cuda:0": 9}
• Prompt/decode: 8 / 1 tokens
• Median TTFT: 246.2100 s
• Median prefill: 0.049 tok/s
• Median decode: 0.024 tok/s
• Cold/warm load: 135.157 s / not measured (memory-safe single-load mode)
• Peak physical device memory: 11.669 GiB
• PyTorch logical allocator peak (diagnostic, may exceed physical VRAM under WSL): 10.920 GiB
• Peak build RAM / VRAM: 29.428 / 1.317 GiB

Limitations

The RTX 4070 validation may use CPU offload and is not a pure-GPU throughput claim unless the performance mode explicitly says so. Preserved MTP weights do not establish speculative decoding support. Compatibility and quality are established only for the exact revision and runtime above. Machine-readable structural, generation, vision, tools, numerical, and performance reports are stored under validation/ in this repository. Remote verification evidence is retained by the build worker and tied to the exact repository revision.