--- title: Qwen2.5-7B-Instruct-FP8-A16 canonical_url: "https://www.modelscope.cn/models/okwinds/Qwen2.5-7B-Instruct-FP8-A16" md_url: "https://www.modelscope.cn/models/okwinds/Qwen2.5-7B-Instruct-FP8-A16.md" repository: okwinds/Qwen2.5-7B-Instruct-FP8-A16 chinese_name: "通义千问Qwen2.5-7B-Instruct-FP8-A16量化高精校准" last_updated: 2024-12-01 license: qwen pipeline_tag: text-generation tasks: - text-generation model_type: - qwen2 architectures: - Qwen2ForCausalLM base_model: - Qwen/Qwen2.5-7B-Instruct base_model_relation: quantized parameters: 7.6B tensor_type: - BF16 - F8_E4M3 - F32 library_name: - safetensors - pytorch frameworks: - pytorch language: - en - zh inference_backends: - "deploy_task text/emb" - "lmdeploy 0.9.1" - "lmdeploy_turbomind 0.9.1" - "sglang 0.5.2" - "vllm 0.9.2" downloads: 513 stars: 5 tags: - Instruct - Chat --- # Qwen2.5-7B-Instruct-FP8-A16 > Qwen2.5-7B-Instruct-FP8-A16 - okwinds 在 ModelScope 开源的模型。通义千问2.5-7B-Instruct-FP8-A16量化高精校准 okwinds/Qwen2.5-7B-Instruct-FP8-A16 是 ModelScope 魔搭社区上的 7.6B 参数text-generation模型，采用 qwen 许可，基于 Qwen/Qwen2.5-7B-Instruct 构建，可用 deploy_task text/emb、lmdeploy 0.9.1、lmdeploy_turbomind 0.9.1 部署。 - **Repository**: okwinds/Qwen2.5-7B-Instruct-FP8-A16 - **License**: qwen - **Tasks**: text-generation - **Parameters**: 7.6B - **Base model**: Qwen/Qwen2.5-7B-Instruct - **Inference backends**: deploy_task text/emb, lmdeploy 0.9.1, lmdeploy_turbomind 0.9.1, sglang 0.5.2, vllm 0.9.2 - **Tags**: Instruct, Chat - **Downloads**: 513 - **Stars**: 5 - **Last updated**: 2024-12-01 Source: https://www.modelscope.cn/models/okwinds/Qwen2.5-7B-Instruct-FP8-A16 --- # 通义千问2.5-7B-Instruct-FP8-A16量化高精校准 原模型 [qwen/Qwen2.5-7B-Instruct](https://www.modelscope.cn/models/qwen/Qwen2.5-7B-Instruct) ```python # download model from modelscope import snapshot_download model_dir = snapshot_download('okwinds/Qwen2.5-7B-Instruct-FP8-A16') ``` ## Qwen2.5 同系列量化模型 | **Model** | GPTQ Compression & FP8 | GGUF | **MLX** (macOS) | | ------------ |----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| -------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | *3B params* | - | [3B-GGUF-V3-LOT](https://www.modelscope.cn/models/okwinds/Qwen2.5-3B-Instruct-GGUF-V3-LOT) | [3B-MLX-4bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-3B-Instruct-MLX-4bit)
[3B-MLX-8bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-3B-Instruct-MLX-8bit) | | *7B params* | [7B-Int4-W4A16](https://www.modelscope.cn/models/okwinds/Qwen2.5-7B-Instruct-Int4-W4A16)
[7B-Int8-W8A16](https://www.modelscope.cn/models/okwinds/Qwen2.5-7B-Instruct-Int8-W8A16)
[7B-FP8-A16](https://www.modelscope.cn/models/okwinds/Qwen2.5-7B-Instruct-FP8-A16)
[Coder-7B-Instruct-FP8](https://www.modelscope.cn/models/okwinds/Qwen2.5-Coder-7B-Instruct-FP8) | [7B-GGUF-V3-LOT](https://www.modelscope.cn/models/okwinds/Qwen2.5-7B-Instruct-GGUF-V3-LOT) | [7B-MLX-4bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-7B-Instruct-MLX-4bit)
[7B-MLX-8bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-7B-Instruct-MLX-8bit) | | *Coder 7B params* | [Coder-7B-Int4-W4A16](https://www.modelscope.cn/models/okwinds/Qwen2.5-Coder-7B-Instruct-Int4-W4A16)
[Coder-7B-Int8-W8A16](https://www.modelscope.cn/models/okwinds/Qwen2.5-Coder-7B-Instruct-Int8-W8A16)
[Coder-7B-FP8-A16](https://www.modelscope.cn/models/okwinds/Qwen2.5-Coder-7B-Instruct-FP8-A16) | [Coder-7B-GGUF-V3-LOT](https://www.modelscope.cn/models/okwinds/Qwen2.5-Coder-7B-Instruct-GGUF-V3-LOT) | [Coder-7B-MLX-4bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-Coder-7B-Instruct-MLX-4bit)
[Coder-7B-MLX-8bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-Coder-7B-Instruct-MLX-8bit) | | *14B params* | [14B-Int4-W4A16](https://www.modelscope.cn/models/okwinds/Qwen2.5-14B-Instruct-Int4-W4A16)
[14B-Int8-W8A16](https://www.modelscope.cn/models/okwinds/Qwen2.5-14B-Instruct-Int8-W8A16)
[14B-FP8-A16](https://www.modelscope.cn/models/okwinds/Qwen2.5-14B-Instruct-FP8-A16) | [14B-GGUF-V3-LOT](https://www.modelscope.cn/models/okwinds/Qwen2.5-14B-Instruct-GGUF-V3-LOT) | [14B-MLX-4bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-14B-Instruct-MLX-4bit)
[14B-MLX-8bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-14B-Instruct-MLX-8bit) | | *Coder 14B params* | [Coder-14B-Int4-W4A16](https://www.modelscope.cn/models/okwinds/Qwen2.5-Coder-14B-Instruct-Int4-W4A16)
[Coder-14B-Int8-W8A16](https://www.modelscope.cn/models/okwinds/Qwen2.5-Coder-14B-Instruct-Int8-W8A16) | [Coder-14B-GGUF-V3-LOT](https://www.modelscope.cn/models/okwinds/Qwen2.5-Coder-14B-Instruct-GGUF-V3-LOT) | [Coder-14B-MLX-4bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-Coder-14B-Instruct-MLX-4bit)
[Coder-14B-MLX-8bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-Coder-14B-Instruct-MLX-8bit) | | *32B params* | [32B-Int4-W4A16](https://www.modelscope.cn/models/okwinds/Qwen2.5-32B-Instruct-Int4-W4A16)
[32B-Int8-W8A16](https://www.modelscope.cn/models/okwinds/Qwen2.5-32B-Instruct-Int8-W8A16)
[32B-FP8-A16](https://www.modelscope.cn/models/okwinds/Qwen2.5-32B-Instruct-FP8-A16) | [32B-GGUF-V3-LOT](https://www.modelscope.cn/models/okwinds/Qwen2.5-32B-Instruct-GGUF-V3-LOT) | [32B-MLX-2bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-32B-Instruct-MLX-2bit)
[32B-MLX-4bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-32B-Instruct-MLX-4bit)
[32B-MLX-8bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-32B-Instruct-MLX-8bit) | | *Coder 32B params* | [Coder-32B-Int4-W4A16](https://www.modelscope.cn/models/okwinds/Qwen2.5-Coder-32B-Instruct-Int4-W4A16)
[Coder-32B-FP8](https://www.modelscope.cn/models/okwinds/Qwen2.5-Coder-32B-Instruct-FP8) | [Coder-32B-GGUF-V3-LOT](https://www.modelscope.cn/models/okwinds/Qwen2.5-Coder-32B-Instruct-GGUF-V3-LOT) | [Coder-32B-MLX-4bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-Coder-32B-Instruct-MLX-4bit)
[Coder-32B-MLX-8bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-Coder-32B-Instruct-MLX-8bit) | | *QwQ-Preview 32B params* | [QwQ-32B-Preview-Int4-W4A16](https://www.modelscope.cn/models/okwinds/QwQ-32B-Preview-Int4-W4A16)
[QwQ-32B-Preview-FP8](https://www.modelscope.cn/models/okwinds/QwQ-32B-Preview-FP8) | [QwQ-32B-GGUF-V3-LOT](https://www.modelscope.cn/models/okwinds/QwQ-32B-Preview-GGUF-V3-LOT) | [QwQ-32B-MLX-4bit](https://www.modelscope.cn/models/okwinds/QwQ-32B-Preview-MLX-4bit)
[QwQ-32B-MLX-8bit](https://www.modelscope.cn/models/okwinds/QwQ-32B-Preview-MLX-8bit) | | *72B params* | - | [72B-GGUF-V3-LOT](https://www.modelscope.cn/models/okwinds/Qwen2.5-72B-Instruct-GGUF-V3-LOT) | [72B-MLX-2bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-72B-Instruct-MLX-2bit)
[72B-MLX-4bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-72B-Instruct-MLX-4bit)
[72B-MLX-8bit](https://www.modelscope.cn/models/okwinds/Qwen2.5-72B-Instruct-MLX-8bit) | ## 模型概述 > 本模型已升级，请使用精度更高的最新版本，原版本归档至 tag v1.0.0。
更多精度详情，请关注 Benchmark 部分数据。 通义千问2.5-7B-Instruct-FP8-A16 是一个基于 通义千问2.5-7B-Instruct 的 FP8 量化并校准的模型。 - **模型名称:** Qwen2.5-7B-Instruct-FP8-A16 - **模型架构:** Qwen2.5 - **权重量化:** FP8 ### 量化模型 该模型通过将 [Qwen2.5-7B-Instruct](https://www.modelscope.cn/models/qwen/Qwen2.5-7B-Instruct) 的权重量化为 FP8 数据类型而实现。 量化过程将每个参数从 16bit 减少到 8bit，将模型占用磁盘空间大小，以及推理时加载模型需要的GPU显存空间，减少到了大约为原模型的1/2。 FP8（Floating Point 8-bit）是一种8位浮点数数据格式，由Nvidia、Arm、Intel联合推出，旨在加速深度学习的训练和推理过程。 本模型的量化，采用 F8_E4M3 数据类型仅对 transformer 中的 Linear 层权重进行 static 量化，其他层均保持为原模型数据类型，采用混合精度的计算方式，尽可能减少模型量化后所带来的精度损失。 并且，在量化过程中，进行了数据校准，以进一步提升模型生成精度。 > 注意：本模型需要 compute capability > 8.0（Ampere、Ada Lovelace、Hopper 架构）的 Nvidia GPU 来支持 FP8 及**混合精度**计算。 ## 部署推理 #### 推荐使用 vLLM （ version > 0.5.0 ） Openai api 兼容模式 ```bash >>> vllm serve "/home/gavin/llm/Qwen2.5-7B-Instruct-FP8-A16" --host 0.0.0.0 --port 8000 --quantization fp8 --gpu-memory-utilization 0.9 --served-model-name "Qwen2.5-7B-Instruct-FP8-A16" ``` ## Benchmark _FP8 (A16) 的量化方法，对比原模型在精度上的表现，几乎无损失。在做数据校准的时候，也最大限度的保有了BF16原模型的精度。_ 本模型的最新版本对精度进行了更广泛的强化提升。 | Benchmark | Qwen2.5-7B-Instruct | FP8-A16-v1.0.0 | Recovery-v1.0.0 | _FP8-A16-master(New)_ | Recovery-master(New) | | -------------------- | ------------------- | -------------- | --------------- | -------------------------------------------------------------------- | ------------------------------------------------------------------- | | **Overall** | 70.69 | 70.94 | 100.36% | 70.98 | 100.41% | | Exam | 77.90 | 77.48 | 99.46% | 77.77 | 99.84% | | Language | 53.97 | 53.84 | 99.75% | 54.13 | 100.29% | | Knowledge | 47.07 | 47.36 | 100.62% | 47.90 | 101.76% | | Understanding | 89.74 | 89.46 | 99.69% | 89.70 | 99.96% | | Coding | 79.32 | 80.68 | 101.71% | 80.45 | 101.42% | | Reasoning | 69.22 | 68.95 | 99.62% | 69.00 | 99.69% | | Instruct & Tools | 77.59 | 78.81 | 101.57% | 77.97 | 100.49% | | **Exam** | - | - | - | - | - | | C-EVAL | 78.52 | 78.15 | 99.53% | 78.29 | 99.71% | | MMLU | 74.19 | 73.80 | 99.47% | 74.01 | 99.76% | | CMMLU | 78.78 | 78.49 | 99.63% | 78.57 | 99.73% | | GaokaoBench | 80.10 | 79.46 | 99.20% | 80.19 | 100.11% | | **Language** | - | - | - | - | - | | WiC | 55.80 | 54.86 | 98.32% | 55.96 | 100.29% | | WSC | 70.19 | 71.15 | 101.37% | 71.15 | 101.37% | | Afqmc-dev | 70.64 | 70.00 | 99.09% | 70.23 | 99.42% | | Tydiqa Goldp | 19.26 | 19.33 | 100.36% | 19.19 | 99.64% | | **Knowledge** | - | - | - | - | - | | BoolQ | 84.92 | 85.26 | 100.40% | 85.47 | 100.65% | | TriviaQA | 53.42 | 52.67 | 98.60% | 52.86 | 98.95% | | GPQA Diamond | 31.82 | 33.84 | 106.35% | 35.35 | 111.09% | | Natual Questions | 18.12 | 17.67 | 97.52% | 17.92 | 98.90% | | **Understanding** | - | - | - | - | - | | C3 | 91.51 | 91.62 | 100.12% | 91.56 | 100.05% | | Race-middle | 91.09 | 90.32 | 99.15% | 90.60 | 99.46% | | Race-high | 86.62 | 86.45 | 99.80% | 86.91 | 100.33% | | **Coding** | - | - | - | - | - | | Human Eval | 83.54 | 83.54 | 100.00% | 85.98 | 102.92% | | Sanitized MBPP | 75.10 | 77.82 | 103.62% | 74.91 | 99.75% | | **Reasoning** | - | - | - | - | - | | COPA | 96.00 | 96.00 | 100.00% | 96.00 | 100.00% | | MATH | 54.12 | 54.96 | 101.55% | 54.72 | 101.11% | | GSM8K | 87.26 | 86.73 | 99.39% | 86.96 | 99.66% | | Winogrande | 72.14 | 71.19 | 98.68% | 71.72 | 99.42% | | Hellaswag | 85.38 | 85.04 | 99.60% | 84.96 | 99.51% | | BBH | 70.12 | 70.26 | 100.20% | 70.26 | 100.20% | | TheoremQA | 19.50 | 18.50 | 94.87% | 18.35 | 94.10% | | **Instruct & Tools** | - | - | - | - | - | | IFEval | 78.66 | 81.41 | 103.50% | 79.62 | 101.22% | | TEval | 76.52 | 76.20 | 99.58% | 76.31 | 99.73% | _以上 benchmark 以及量化工作均在 Nvidia 4090 RTX 24GB 显卡上完成。_ ***并发说明*** 在以上 benchmark 基准测试中，采用 vllm 推理，观察到以下基准测试中的并发表现。 - vLLM 0.5.5 - 原模型 BF16 峰值并发，在无 pending 的前提下，达到 running 90 左右。 - 量化后的本模型 FP8(A16) 峰值并发，在无 pending 的前提下，达到 running 140 左右。 - vLLM 0.6.1.post2 - 原模型 BF16 峰值并发，在无 pending 的前提下，达到 running 130 左右。 - 量化后的本模型 FP8(A16) 峰值并发，在无 pending 的前提下，达到 running 250 左右。