--- title: GLM-5.2-FP8-DFlash canonical_url: "https://www.modelscope.cn/models/UCloud-AILab/GLM-5.2-FP8-DFlash" md_url: "https://www.modelscope.cn/models/UCloud-AILab/GLM-5.2-FP8-DFlash.md" repository: UCloud-AILab/GLM-5.2-FP8-DFlash last_updated: 2026-07-11 license: mit pipeline_tag: text-generation tasks: - text-generation model_type: - qwen3 architectures: - DFlashDraftModel base_model: - ZhipuAI/GLM-5.2-FP8 base_model_relation: finetune parameters: 3.7B tensor_type: - BF16 library_name: - safetensors downloads: 102 stars: 2 tags: - dflash - speculative-decoding - draft-model - block-diffusion - glm --- # GLM-5.2-FP8-DFlash > GLM-5.2-FP8-DFlash - UCloud-AILab 在 ModelScope 开源的模型。Paper | DFlash GitHub | SpecForge UCloud-AILab/GLM-5.2-FP8-DFlash 是 ModelScope 魔搭社区上的 3.7B 参数text-generation模型，采用 mit 许可，基于 ZhipuAI/GLM-5.2-FP8 构建。 - **Repository**: UCloud-AILab/GLM-5.2-FP8-DFlash - **License**: mit - **Tasks**: text-generation - **Parameters**: 3.7B - **Base model**: ZhipuAI/GLM-5.2-FP8 - **Tags**: dflash, speculative-decoding, draft-model, block-diffusion, glm - **Downloads**: 102 - **Stars**: 2 - **Last updated**: 2026-07-11 Source: https://www.modelscope.cn/models/UCloud-AILab/GLM-5.2-FP8-DFlash --- # GLM-5.2-FP8-DFlash [Paper](https://arxiv.org/abs/2602.06036) | [DFlash GitHub](https://github.com/z-lab/dflash) | [SpecForge](https://github.com/sgl-project/SpecForge) DFlash block-diffusion speculative-decoding drafter for [GLM-5.2-FP8](https://modelscope.cn/models/ZhipuAI/GLM-5.2-FP8) (743B MoE, 39B active). Standard DFlash method (no extensions), trained with [SpecForge](https://github.com/sgl-project/SpecForge) on the paper-specified data recipe: ~800K samples of Nemotron-Post-Training-v2 + CodeAlpaca (code / math / chat), all responses regenerated by GLM-5.2-FP8, 6 epochs — directly comparable to the DFlash paper and the z-lab drafter series. ## Quick Start (SGLang) ```bash SGLANG_USE_MODELSCOPE=true python -m sglang.launch_server \ --model-path zai-org/GLM-5.2-FP8 \ --speculative-algorithm DFLASH \ --speculative-draft-model-path UCloud-AILab/GLM-5.2-FP8-DFlash \ --speculative-num-draft-tokens 16 \ --tp-size 8 \ --trust-remote-code ``` vLLM v0.20.1+ has native DFlash support and reads this checkpoint directly, no conversion needed. Not yet runtime-verified on our infrastructure; the results below were produced via SGLang. ## Evaluation ### Mean accepted length & end-to-end speedup Measured on live SGLang serving (concurrency 1, greedy decoding unless noted). | Benchmark | AL (built-in MTP) | AL (DFlash) | DFlash throughput (tok/s) | Speedup vs vanilla | vs built-in MTP | |-----------|-------------------|-------------|---------------------------|--------------------|-----------------| | gsm8k | 4.01 | 4.44 | 236 | 2.22x | 1.51x | | humaneval | 4.42 | 6.43 | 383 | 3.44x | 1.42x | | math500 | 4.91 | 7.77 | 477 | 4.28x | 1.54x | | mbpp | 5.23 | 8.04 | 478 | 4.29x | 1.46x | | mtbench | 3.71 | 3.56 | 220 | 1.99x | 0.93x | | ceval | 3.65 | 2.98 | 177 | 1.62x | 0.93x | Built-in MTP baseline uses the official GLM-5.2 recipe (EAGLE, steps 5 / topk 1 / draft tokens 6). This drafter is code/math-optimized: it delivers 1.4-1.5x over the (already strong) built-in MTP on code and math workloads, while chat and Chinese-language workloads slightly favor built-in MTP (see Limitations). ceval (Chinese) is the weakest domain — the training corpus is English-dominant (see Limitations). Also mirrored on HuggingFace: [UCloud-org/GLM-5.2-FP8-DFlash](https://huggingface.co/UCloud-org/GLM-5.2-FP8-DFlash). ## Training Details - Target model: GLM-5.2-FP8 (hidden 6144, 78 layers; drafter conditions on target layers [1, 20, 38, 56, 75]) - Drafter: 5-layer block-diffusion transformer, block_size 16, **3.7B** total parameters (**1.8B independently trained**; embed/lm_head reused from target, frozen, not trained, included for standalone inference) - Data: [Jessiewei7/GLM-5.2-FP8-nemotron-codealpaca](https://modelscope.cn/datasets/Jessiewei7/GLM-5.2-FP8-nemotron-codealpaca) — Nemotron-Post-Training-v2 + CodeAlpaca, ~800K samples (paper-specified recipe), all responses regenerated by GLM-5.2-FP8 (non-thinking mode), max_length 3072 - Recipe: 6 epochs, AdamW with update clipping (StableAdamW-style), lr 6e-4 cosine (4% warmup), grad-clip 1.0, num_anchors 512, loss_decay_gamma 7, pure cross-entropy (standard DFlash loss) - Framework: SpecForge (offline hidden-state pipeline), FSDP2 ## Limitations - Trained on non-thinking-mode regenerated data; speedup under thinking-mode inference has not been evaluated yet. - Trained with `max_length` 3072: accept length degrades on very long prompts. For long-context workloads, benchmark against the built-in MTP before choosing. - Training corpus is English-dominant: acceptance length on Chinese-language workloads is lower (ceval AL 2.98 vs 4.4-8.0 on English benchmarks). ## Acknowledgements [DFlash](https://github.com/z-lab/dflash) (z-lab), [SpecForge](https://github.com/sgl-project/SpecForge) / [SGLang](https://github.com/sgl-project/sglang) (sgl-project). GLM-5.2 by Zhipu AI. ## Citation If you use this model, please cite: ```bibtex @misc{ucloud2026glm52dflash, title = {GLM-5.2-FP8-DFlash: A DFlash Speculative-Decoding Drafter for GLM-5.2-FP8}, author = {Wei, Xiaojun and {UCloud AILab}}, year = {2026}, howpublished = {\url{https://huggingface.co/UCloud-org/GLM-5.2-FP8-DFlash}} } ``` This model is trained with the DFlash method — please also cite: ```bibtex @misc{chen2026dflash, title = {DFlash: Block Diffusion for Flash Speculative Decoding}, author = {Chen, Jian and Liang, Yesheng and Liu, Zhijian}, year = {2026}, eprint = {2602.06036}, archivePrefix = {arXiv}, primaryClass = {cs.CL}, url = {https://arxiv.org/abs/2602.06036} } ```