MODEL PROFILE

Qwen3.8-27B-FP8-NInfer-v3

yyrgzn · Qwen3.8-27B-FP8-NInfer-v3

This repository contains `Qwen3.8-27B-FP8-v3-full.ninfer`, the full NInfer v3 artifact for Qwen3.8-27B built from the official Qwen/Qwen3.8-27B-FP8 block-wise FP8 checkpoint. The artifact is intended only for NInfer; it is not a Transformers checkpoint, Safetensors distribution, or GGUF file. The Text tower keeps the official E4M3 quantization. The checkpoint stores the linear projections as E4M3 codes with a BF16 `weight_scale_inv` grid per 128×128 weight block; the converter decodes each block exactly (an E4M3 mantissa times a BF16 scale fits FP32 without rounding) and re-encodes the values with one BF16 scale per output row (`fp8_row_maxabs`). The token embedding, output head, and every text projection except the GDN `a`/`b` gating projections (kept direct BF16) use that row-scaled FP8, which the engine runs on its FP8-activation routes. MTP uses Q8 group-32, Vision is full BF16, and DFlash2 uses Q4/Q5/Q6 group-64. The official card reports the block-wise FP8 checkpoint's performance as nearly identical to the BF16 original. The file contains Text, Vision, MTP, DFlash2 and the frontend resources. Vision weights come from the checkpoint's own full-BF16 visual tower. Vision and speculative weights are loaded only when selected at startup.

Visual QAmultimodallicense:apache-2.0
Context windowNot disclosed
Input → OutputText、Image → Text
Model weightsNot verified
Profile updated2026/11/2 19:16:39
OVERVIEW

Model overview

This repository contains `Qwen3.8-27B-FP8-v3-full.ninfer`, the full NInfer v3 artifact for Qwen3.8-27B built from the official Qwen/Qwen3.8-27B-FP8 block-wise FP8 checkpoint. The artifact is intended only for NInfer; it is not a Transformers checkpoint, Safetensors distribution, or GGUF file. The Text tower keeps the official E4M3 quantization. The checkpoint stores the linear projections as E4M3 codes with a BF16 `weight_scale_inv` grid per 128×128 weight block; the converter decodes each block exactly (an E4M3 mantissa times a BF16 scale fits FP32 without rounding) and re-encodes the values with one BF16 scale per output row (`fp8_row_maxabs`). The token embedding, output head, and every text projection except the GDN `a`/`b` gating projections (kept direct BF16) use that row-scaled FP8, which the engine runs on its FP8-activation routes. MTP uses Q8 group-32, Vision is full BF16, and DFlash2 uses Q4/Q5/Q6 group-64. The official card reports the block-wise FP8 checkpoint's performance as nearly identical to the BF16 original. The file contains Text, Vision, MTP, DFlash2 and the frontend resources. Vision weights come from the checkpoint's own full-BF16 visual tower. Vision and speculative weights are loaded only when selected at startup.

Key features

特点摘要待依据来源补齐,不凭模型名称推断能力。

Limitations & considerations

Sources are insufficient to establish specific limitations. Verify task performance, latency, data policies, licensing and version lifecycle.

PRICING & COST

Source prices & cost analysis

Prices vary by channel, region and configuration. Channel quotes are listed independently and are not treated as a single official provider price.

Source / channelInputOutputBilling unitCollected at
No verified prices yet. Check the source channel.

Estimate monthly base token costs

—

Estimates exclude cache discounts, tiered pricing, multimodal surcharges, web search, taxes and additional reasoning output.

INDEPENDENT BENCHMARKS

Public benchmarks & run configuration

暂无已采集的可核验公开评分。模型未上榜不代表能力较低。

View full indexed rankings →

Frequently asked questions

Do open weights permit free commercial use?

Not necessarily. Check the version license, commercial conditions and usage policies. Self-hosting incurs compute and maintenance costs.

Are these real-time billing prices?

These are timestamped source snapshots, not live billing prices. Check the channel for current versions, regions, discounts and billing methods.

Sources & traceability

许可证:apache-2.0。模型介绍与第三方评测均以原来源的具体版本和方法为准。