Hugging Face's logo

stepfun-ai
/
Step-5-Preview-NVFP4

Safetensors
step5_vl
8-bit precision
modelopt
Model card Files
xet
 Community
1

Upload Step-5-Preview NVFP4 v1 weights

#1
by huangyu-nv - opened
base: refs/heads/main
←
from: refs/pr/1
Discussion Files changed
+658439
-1
This view is limited to 50 files because it contains too many changes. See the raw diff here.
Files changed (50) hide show
  1. LICENSE-CN.txt +26 -0
  2. LICENSE.txt +41 -0
  3. README.md +182 -1
  4. SHA256SUMS +102 -0
  5. chat_template.jinja +97 -0
  6. config.json +1070 -0
  7. hf_quant_config.json +384 -0
  8. model-extra-000.safetensors +3 -0
  9. model-extra-001.safetensors +3 -0
  10. model-layer-000.safetensors +3 -0
  11. model-layer-001.safetensors +3 -0
  12. model-layer-002.safetensors +3 -0
  13. model-layer-003.safetensors +3 -0
  14. model-layer-004.safetensors +3 -0
  15. model-layer-005.safetensors +3 -0
  16. model-layer-006.safetensors +3 -0
  17. model-layer-007.safetensors +3 -0
  18. model-layer-008.safetensors +3 -0
  19. model-layer-009.safetensors +3 -0
  20. model-layer-010.safetensors +3 -0
  21. model-layer-011.safetensors +3 -0
  22. model-layer-012.safetensors +3 -0
  23. model-layer-013.safetensors +3 -0
  24. model-layer-014.safetensors +3 -0
  25. model-layer-015.safetensors +3 -0
  26. model-layer-016.safetensors +3 -0
  27. model-layer-017.safetensors +3 -0
  28. model-layer-018.safetensors +3 -0
  29. model-layer-019.safetensors +3 -0
  30. model-layer-020.safetensors +3 -0
  31. model-layer-021.safetensors +3 -0
  32. model-layer-022.safetensors +3 -0
  33. model-layer-023.safetensors +3 -0
  34. model-layer-024.safetensors +3 -0
  35. model-layer-025.safetensors +3 -0
  36. model-layer-026.safetensors +3 -0
  37. model-layer-027.safetensors +3 -0
  38. model-layer-028.safetensors +3 -0
  39. model-layer-029.safetensors +3 -0
  40. model-layer-030.safetensors +3 -0
  41. model-layer-031.safetensors +3 -0
  42. model-layer-032.safetensors +3 -0
  43. model-layer-033.safetensors +3 -0
  44. model-layer-034.safetensors +3 -0
  45. model-layer-035.safetensors +3 -0
  46. model-layer-036.safetensors +3 -0
  47. model-layer-037.safetensors +3 -0
  48. model-layer-038.safetensors +3 -0
  49. model-layer-039.safetensors +3 -0
  50. model-layer-040.safetensors +3 -0
LICENSE-CN.txt ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ StepFun模型许可证1.0版
2
+
3
+ 版权所有 (c) 2026 StepFun AI
4
+ 1.权利的授予
5
+ 受限于本许可证的条款和条件，StepFun AI特此授予您一项永久的（受第5节约束）、全球性的、非排他的、免交易费的、免许可费的许可，以使用、复制、分发和分许可模型卡（“模型卡”）中所标识的StepFun模型，包括其权重、参数、配置、分词器、模型专用代码及文档（合称“本模型”），并修改本模型及制作其衍生成果。
6
+
7
+ StepFun AI及每个贡献者特此授予您一项永久的（受第5节约束）、全球性的、非排他的、免交易费的、免许可费的专利许可，以制造、委托制造、使用、许诺销售、销售、进口及以其他方式转移本模型和衍生成果，该许可仅及于StepFun AI或贡献者各自有权许可、且为StepFun AI或贡献者所提供形式的本模型或衍生成果所必然侵犯的专利权利要求。如果您对任何实体提起专利诉讼（包括在诉讼中提起交叉诉讼或反诉），主张按照StepFun AI或贡献者所提供形式的本模型或衍生成果构成专利的直接侵权或帮助侵权，则本专利许可自该诉讼提起之日起终止。
8
+
9
+ “衍生成果”系指经微调、蒸馏、合并、量化、剪枝、持续预训练或以其他方式修改而成的、复制或包含本模型实质部分的模型。仅通过API调用本模型的独立应用程序，并不构成衍生成果。在实质程度上基于本模型输出训练而成的模型，视为复制或包含了本模型的实质部分。“贡献者”系指制作衍生成果并在本许可证下分发或以其他方式提供该衍生成果的任何个人或法人实体。
10
+ 2.条件
11
+ （a）再分发。当您分发本模型或衍生成果时，您必须提供本许可证的副本，保留所有版权、专利、商标和权属声明，以显著的声明标明您所作的修改，包含随本模型提供的任何NOTICE文件，并提供模型卡或其链接。NOTICE文件中的内容仅为提供信息，并不修改本许可证。您可以在您分发的衍生成果中添加您自己的权属声明，与本模型的NOTICE文本并列或作为其附件，只要该声明不被解释为修改本许可证。
12
+
13
+ （b）署名。如果在任何自然月内，您或您关联方提供的、使用或包含本模型或衍生成果的任何商业产品或服务的活跃用户超过100,000,000人，或产生的收入超过20,000,000美元（或其他货币的等值金额），则您此后必须在该产品或服务的用户界面（包括“关于”页面）及其用户协议或产品文档中显著展示“Powered by StepFun AI”（或StepFun AI书面指定的模型名称），其显著程度不得低于任何其他第三方组件声明，并在相关文档中保留模型名称及许可证声明。
14
+
15
+ （c）商业许可。在您或您关联方实施下列行为之前，您必须与StepFun AI另行签订商业许可协议：（i）作为向终端用户供应硬件设备（包括但不限于手机、车辆、智能眼镜、可穿戴设备、机器人或智能家居设备）的制造商或品牌所有者，首次将本模型或任何衍生成果的能力作为该等设备或其配套服务的功能进行商业化提供，无论直接提供或通过代表您或您关联方行事的第三方提供，无论部署于设备上、远程部署或部署于随设备供应的基础设施上，亦无论在出厂时交付或此后通过激活、下载或更新交付；或（ii）在您或您任何关联方经营MaaS业务期间，将本模型或任何衍生成果用于任何商业目的，且您及您关联方在任何连续十二个月期间内所有业务的合并收入超过20,000,000美元（或其他货币的等值金额）。
16
+ 本第2（c）节不适用于不向任何第三方提供本模型或衍生成果、或其输出或底层能力的内部使用，亦不适用于通过StepFun AI认证的分发渠道获取的本模型或任何衍生成果在该认证范围内的使用。
17
+ “模型即服务”（MaaS）系指向第三方提供对本模型或衍生成果的推理或微调访问（包括通过API），且该第三方对输入、参数或训练数据拥有实质性控制。将本模型的能力嵌入特定终端用户功能而不向第三方提供该等控制，或仅将请求转发至其他提供方的模型，均不构成MaaS。
18
+ 3.合规与商标
19
+ 您应自行负责遵守所有适用的法律法规，包括关于数据、隐私、网络安全、安全、出口管制、制裁、消费者保护及特定行业合规的法律法规，并在您使用、修改或分发本模型或任何衍生成果的过程中尊重第三方权利。本许可证不授予使用StepFun AI的商号、商标、服务标志或产品名称的许可，但在描述本模型来源时的合理和习惯用法以及第2（b）节所要求的有限展示除外。您不得声明或暗示StepFun AI为您或您的产品或服务背书。
20
+ 4.免责声明与责任限制
21
+ 除非所适用法律要求或经书面同意：��A）本模型及任何衍生成果均“按原样”提供，没有任何明示或默示的保证或条件，包括关于所有权、不侵权、适销性或特定目的适用性的任何保证或条件，以及关于任何输出的准确性、可靠性或质量的任何保证或条件，您自行负责确认使用或再分发本模型或任何衍生成果的适当性，并承担与您行使本许可证下所授许可相关的任何风险；（B）在任何情况和任何法律理论下，无论在侵权（包括过失）、合同或其他方面，StepFun AI、其关联方或任何贡献者均不对您的损失承担责任，包括因本许可证或因使用或无法使用本模型、任何衍生成果或任何输出而导致的任何直接的、间接的、特殊的、偶然的或继发性的损失。
22
+ 5.终止
23
+ 对本许可证的任何重大违反，将终止本许可证项下授予您的权利。如您在StepFun AI发出书面通知后三十（30）日内纠正该违反行为，或StepFun AI书面确认恢复，则您的权利将得以恢复。终止后，您必须立即停止使用并销毁您所持有的本模型及任何衍生成果的所有副本，但在终止前合法获取副本的合规接收者可继续行使其在本许可证下的权利。
24
+ 6.其他
25
+ 本许可证的英文版本优先于任何译本。如对本许可证有任何疑问，请联系license@stepfun.ai。
26
+ 条款和条件结束
LICENSE.txt ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ StepFun Model License 1.0
2
+
3
+ Copyright (c) 2026 StepFun AI
4
+
5
+ 1. Grant of Rights
6
+
7
+ Subject to the terms and conditions of this License, StepFun AI hereby grants You a perpetual (subject to Section 5), worldwide, non-exclusive, no-charge, royalty-free license to use, copy, distribute, and sublicense the StepFun model identified in the model card (the "Model Card"), including its weights, parameters, configuration, tokenizer, model-specific code, and documentation (collectively, the "Model"), and to modify the Model and create Derivative Works thereof.
8
+
9
+ StepFun AI and each Contributor hereby grant You a perpetual (subject to Section 5), worldwide, non-exclusive, no-charge, royalty-free patent license to make, have made, use, offer to sell, sell, import, and otherwise transfer the Model and Derivative Works, where such license applies only to those patent claims licensable by StepFun AI or such Contributor, respectively, that are necessarily infringed by the Model or a Derivative Work, in each case in the form provided by StepFun AI or such Contributor. This patent license terminates as of the date You institute patent litigation against any entity (including a cross-claim or counterclaim in a lawsuit) alleging that the Model or a Derivative Work in the form provided by StepFun AI or such Contributor constitutes direct or contributory patent infringement.
10
+
11
+ "Derivative Work" means a fine-tuned, distilled, merged, quantized, pruned, continually pre-trained, or otherwise modified model that reproduces or contains a substantial part of the Model. A separate application that only calls the Model through an API is not a Derivative Work solely for that reason. A model trained to a material extent on outputs of the Model is deemed to reproduce or contain a substantial part of the Model. "Contributor" means any individual or legal entity that creates a Derivative Work and distributes or otherwise makes it available under this License.
12
+
13
+ 2. Conditions
14
+
15
+ (a) Redistribution. When You distribute the Model or a Derivative Work, You must provide a copy of this License, retain all copyright, patent, trademark, and attribution notices, identify Your modifications with prominent notices, include any NOTICE file supplied with the Model, and provide the Model Card or a link to it. The contents of any NOTICE file are for informational purposes only and do not modify this License. You may add Your own attribution notices within Derivative Works that You distribute, alongside or as an addendum to the NOTICE text from the Model, provided that such notices cannot be construed as modifying this License.
16
+
17
+ (b) Attribution. If, in any calendar month, any commercial product or service provided by You or Your affiliates that uses or incorporates the Model or a Derivative Work has more than 100,000,000 active users or generates revenue exceeding US$20,000,000 (or the equivalent in another currency), You must thereafter prominently display "Powered by StepFun AI" (or such Model name as StepFun AI designates in writing) on the user interface of that product or service (including the "About" page) and in its user agreement or product documentation with prominence at least equal to that of any other third-party component notice, and retain the Model name and license notices in related documentation.
18
+
19
+ (c) Commercial License. You must enter into a separate commercial license agreement with StepFun AI before You or Your affiliates: (i) acting as a manufacturer or brand owner of hardware devices (including, without limitation, phones, vehicles, smart glasses, wearables, robots, or smart-home devices) supplied to end users, first make the capabilities of the Model or any Derivative Work commercially available as a feature of those devices or their accompanying services, whether directly or through third parties acting on Your or their behalf, whether deployed on the device, remotely, or on infrastructure supplied with the device, and whether delivered at shipment or later by activation, download, or update; or (ii) use the Model or any Derivative Work for any commercial purpose while You or any of Your affiliates operate a MaaS business, if the combined revenue of You and Your affiliates from all sources exceeds US$20,000,000 (or the equivalent in another currency) over any consecutive twelve-month period.
20
+
21
+ This Section 2(c) does not apply to internal use that does not make the Model or a Derivative Work, or the outputs or underlying capabilities thereof, available to any third party, or to use of the Model or any Derivative Work obtained through a distribution channel certified by StepFun AI, within the scope of such certification.
22
+
23
+ "Model as a Service" (MaaS) means giving a third party inference or fine-tuning access to the Model or a Derivative Work, including through an API, where the third party has meaningful control over inputs, parameters, or training data. Neither embedding the Model's capabilities in a specific end-user feature without giving that control to a third party, nor merely relaying requests to another provider's model, constitutes MaaS.
24
+
25
+ 3. Compliance and Trademarks
26
+
27
+ You are solely responsible for complying with all applicable laws and regulations, including those relating to data, privacy, security, safety, export control, sanctions, consumer protection, and sector-specific compliance, and for respecting third-party rights in connection with Your use, modification, or distribution of the Model or any Derivative Work. This License does not grant permission to use the trade names, trademarks, service marks, or product names of StepFun AI, except for reasonable and customary use in describing the origin of the Model and the limited display required by Section 2(b). You must not state or imply that StepFun AI endorses You or Your products or services.
28
+
29
+ 4. Disclaimer and Liability
30
+
31
+ UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING: (A) THE MODEL AND ANY DERIVATIVE WORK ARE PROVIDED "AS IS", WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING ANY WARRANTIES OR CONDITIONS OF TITLE, NON-INFRINGEMENT, MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, OR REGARDING THE ACCURACY, RELIABILITY, OR QUALITY OF ANY OUTPUT, AND YOU ARE SOLELY RESPONSIBLE FOR DETERMINING THE APPROPRIATENESS OF USING OR REDISTRIBUTING THE MODEL OR ANY DERIVATIVE WORK, AND FOR ASSUMING ANY RISKS ASSOCIATED WITH YOUR EXERCISE OF PERMISSIONS UNDER THIS LICENSE; AND (B) IN NO EVENT AND UNDER NO LEGAL THEORY, WHETHER IN TORT (INCLUDING NEGLIGENCE), CONTRACT, OR OTHERWISE, SHALL STEPFUN AI, ITS AFFILIATES, OR ANY CONTRIBUTOR BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES OF ANY CHARACTER ARISING AS A RESULT OF THIS LICENSE OR OUT OF THE USE OR INABILITY TO USE THE MODEL, ANY DERIVATIVE WORK, OR ANY OUTPUT.
32
+
33
+ 5. Termination
34
+
35
+ Any material breach of this License terminates the rights granted to You hereunder. Your rights will be reinstated if the breach is cured within thirty (30) days after written notice from StepFun AI, or if StepFun AI confirms such reinstatement in writing. Upon termination, You must immediately cease all use of, and destroy all copies of, the Model and any Derivative Work in Your possession, except that compliant recipients who lawfully acquired copies thereof prior to termination may continue to exercise their rights under this License.
36
+
37
+ 6. Miscellaneous
38
+
39
+ The English version of this License shall prevail over any translation. For any questions regarding this License, please contact license@stepfun.ai.
40
+
41
+ END OF TERMS AND CONDITIONS
README.md CHANGED
@@ -1,3 +1,184 @@
1
  ---
2
- license: apache-2.0
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ license: other
3
+ license_link: https://huggingface.co/stepfun-ai/Step-5-Preview-NVFP4/blob/main/LICENSE.txt
4
+ base_model:
5
+ - stepfun-ai/Step-5-Preview
6
+ base_model_relation: quantized
7
  ---
8
+
9
+ <p align="center">
10
+ <img src="https://cdn-uploads.huggingface.co/production/uploads/65dc9de42a6348b052a3286e/-R9DERgmQiBfO2cF0HXoD.png" alt="StepFun logo" width="64" />
11
+ </p>
12
+
13
+ <h1 align="center">Step 5 Preview — NVFP4</h1>
14
+
15
+ <p align="center">
16
+ <strong>StepFun's flagship model for agentic work — NVFP4 checkpoint for NVIDIA Blackwell</strong>
17
+ </p>
18
+
19
+ <p align="center">
20
+ <strong>600B total parameters &nbsp;·&nbsp; 27B active parameters &nbsp;·&nbsp; Up to 1M context</strong>
21
+ </p>
22
+
23
+ <p align="center">
24
+ <a href="https://www.stepfun.com/step-5-preview"><img src="https://img.shields.io/badge/Model_Page-00BFA5?style=flat-square" alt="Model Page" /></a>
25
+ <a href="https://platform.stepfun.ai/docs/en/guides/models/step-5-preview"><img src="https://img.shields.io/badge/Documentation-2563EB?style=flat-square" alt="Documentation" /></a>
26
+ <a href="https://platform.stepfun.ai/"><img src="https://img.shields.io/badge/API_Platform-7C3AED?style=flat-square" alt="API Platform" /></a>
27
+ </p>
28
+
29
+ <p align="center">
30
+ <a href="#introduction">Introduction</a> &nbsp;·&nbsp;
31
+ <a href="#key-capabilities">Capabilities</a> &nbsp;·&nbsp;
32
+ <a href="#model-specifications">Specifications</a> &nbsp;·&nbsp;
33
+ <a href="#nvfp4-quantization">NVFP4</a> &nbsp;·&nbsp;
34
+ <a href="#performance">Performance</a> &nbsp;·&nbsp;
35
+ <a href="#deployment">Deployment</a> &nbsp;·&nbsp;
36
+ <a href="#resources">Resources</a>
37
+ </p>
38
+
39
+ ---
40
+
41
+ ## Introduction
42
+
43
+ This repository hosts the **NVFP4 quantization** of Step 5 Preview, built in collaboration with NVIDIA for NVIDIA Blackwell GPUs. For the FP8 release checkpoint and the standalone Transformers reference implementation, see [stepfun-ai/Step-5-Preview](https://huggingface.co/stepfun-ai/Step-5-Preview); for BF16, see [stepfun-ai/Step-5-Preview-BF16](https://huggingface.co/stepfun-ai/Step-5-Preview-BF16).
44
+
45
+ Step 5 Preview is StepFun's flagship model for agentic work. It delivers frontier-level performance across software engineering and professional knowledge work, with particular strength in finance.
46
+
47
+ Step 5 Preview is a sparse Mixture-of-Experts (MoE) model with 600B total parameters and 27B active parameters. It supports up to 1 million tokens of context and accepts text, images, and video, enabling sustained reasoning and tool use across complex tasks.
48
+
49
+ ## Key Capabilities
50
+
51
+ ### Software Engineering
52
+
53
+ Step 5 Preview supports repository-level software engineering, frontend and visual development, and programming for devices. It can navigate large codebases, coordinate changes across files, run and verify code, and iterate on the results.
54
+
55
+ ### Long-Horizon Agentic Execution
56
+
57
+ Step 5 Preview can carry out tasks that require repeated tool calls, execution, verification, and revision, drawing on long-context information throughout the workflow.
58
+
59
+ ### Professional Knowledge Work
60
+
61
+ The model combines research, multimodal understanding, and tool use to analyze long documents, charts, screenshots, images, and video, and produce structured reports and analyses.
62
+
63
+ ### A Focus on Finance
64
+
65
+ Step 5 Preview is particularly strong in financial tasks that combine quantitative reasoning with analysis of financial data and supporting documents.
66
+
67
+ ## Model Specifications
68
+
69
+ | Specification | Step 5 Preview — NVFP4 |
70
+ | --- | --- |
71
+ | Architecture | Sparse Mixture-of-Experts (MoE) |
72
+ | Total Parameters | 600B |
73
+ | Active Parameters | 27B |
74
+ | Context Window | Up to 1M tokens |
75
+ | Maximum Output | Up to 1M tokens |
76
+ | Input | Text, image |
77
+ | Output | Text |
78
+ | Weights | NVFP4 (~395 GB on disk) |
79
+
80
+ ## NVFP4 Quantization
81
+
82
+ This checkpoint was quantized with the [NVIDIA TensorRT Model Optimizer](https://github.com/NVIDIA/TensorRT-Model-Optimizer) (ModelOpt 0.47).
83
+
84
+ - **Format.** NVFP4 is a 4-bit floating-point format with two-level microscaling: an FP8 scale per 16-element group plus a per-tensor scale. Linear and MoE expert weights are stored in NVFP4.
85
+ - **Precision-sensitive modules are preserved.** Embeddings, `lm_head`, final norm, the early dense layers, attention projections, shared experts, and MoE router gates remain in the original precision. The KV cache is not quantized.
86
+ - **Footprint.** Roughly 35% smaller than the FP8 checkpoint (~395 GB vs. ~600 GB), leaving more HBM for KV cache at long context.
87
+ - **Hardware.** NVFP4 GEMMs require NVIDIA Blackwell GPUs (B200 / GB200 / GB300 / RTX PRO 6000 Blackwell). Serving is supported through vLLM.
88
+
89
+ Accuracy was validated against the FP8 release checkpoint across reasoning, coding, agentic, and multimodal benchmarks.
90
+
91
+ ## Performance
92
+
93
+ The following results summarize Step 5 Preview's performance across coding, agentic tasks, professional knowledge work, and finance. Scores shown are for the Step 5 Preview release checkpoint.
94
+
95
+ <p align="center">
96
+ <img src="https://cdn-uploads.huggingface.co/production/uploads/65dc9de42a6348b052a3286e/mfY2eltwq9TlXWlK6lx7K.png" alt="Step 5 Preview benchmark results across coding, agentic tasks, professional knowledge work, and finance" width="100%" />
97
+ </p>
98
+
99
+ ### Evaluation Notes
100
+
101
+ - Recommended inference settings: `temperature=1.0`, `top_p=0.95`, `top_k=0`.
102
+ - Step 5 Preview results shown above use the **High** reasoning setting.
103
+ - GDPval-AA v2.1 scores are based on results from [Artificial Analysis](https://artificialanalysis.ai/) as of September 20, 2026.
104
+ - DeepSWE v1.1 evaluations use the [SWE-agent](https://github.com/SWE-agent/SWE-agent) harness with `temperature=1.0` and `top_p=0.95`.
105
+
106
+ For evaluation context, methodology, and detailed case studies, see the [Step 5 Preview Model Page](https://www.stepfun.com/step-5-preview).
107
+
108
+ ## Deployment
109
+
110
+ The NVFP4 checkpoint is served through an OpenAI-compatible API with vLLM on NVIDIA Blackwell GPUs. For FP8 and BF16 deployment (vLLM, SGLang, and the standalone Transformers implementation), see [stepfun-ai/Step-5-Preview](https://huggingface.co/stepfun-ai/Step-5-Preview).
111
+
112
+ Use a framework build that includes Step 5 Preview support, including the reasoning, tool-call, and speculative decoding features used in the serving commands.
113
+
114
+ ### Deployment at a Glance
115
+
116
+ | Framework | Precision | Hardware | Parallelism | Interface |
117
+ | --- | --- | --- | --- | --- |
118
+ | vLLM | NVFP4 | 4× B200-class GPUs (one GB200 node) | TP4 | OpenAI-compatible API |
119
+ | vLLM | NVFP4 | 8× Blackwell GPUs | TP8 | OpenAI-compatible API |
120
+
121
+ These example configurations are sized for the deployments below. Actual memory requirements, usable context length, and concurrency vary with the inference framework and KV cache allocation; with more aggregate HBM (e.g., TP8), longer context windows are possible.
122
+
123
+ ### vLLM
124
+
125
+ #### Environment Setup
126
+
127
+ We recommend using the official vLLM Docker image:
128
+
129
+ ```bash
130
+ docker pull vllm/vllm-openai:latest
131
+ ```
132
+
133
+ After pulling the image, follow the [vLLM Docker guide](https://docs.vllm.ai/en/latest/deployment/docker/) to start a GPU-enabled container with access to your model weights, then run the command below inside it.
134
+
135
+ <details>
136
+ <summary><strong>NVFP4: 4× B200-class GPUs, TP4</strong></summary>
137
+
138
+ To serve the NVFP4 checkpoint on four Blackwell GPUs with 256K context, run the following command. It serves the model as `step5`:
139
+
140
+ ```bash
141
+ vllm serve stepfun-ai/Step-5-Preview-NVFP4 \
142
+ --served-model-name "step5" \
143
+ --tensor-parallel-size 4 \
144
+ --max-model-len 262144 \
145
+ --max-num-batched-tokens 16384 \
146
+ --max-num-seqs 32 \
147
+ --gpu-memory-utilization 0.90 \
148
+ --block-size 64 \
149
+ --enable-prefix-caching \
150
+ --spec-method mtp \
151
+ --spec-tokens 3 \
152
+ --reasoning-parser step5 \
153
+ --enable-auto-tool-choice \
154
+ --tool-call-parser step5
155
+ ```
156
+
157
+ For eight GPUs, set `--tensor-parallel-size 8` and increase `--max-model-len` as HBM allows.
158
+
159
+ </details>
160
+
161
+ ### Send a Request
162
+
163
+ Once the server is ready:
164
+
165
+ ```bash
166
+ export BASE_URL="http://localhost:8000"
167
+ curl "$BASE_URL/v1/chat/completions" \
168
+ -H "Content-Type: application/json" \
169
+ -d '{
170
+ "model": "step5",
171
+ "messages": [
172
+ {"role": "user", "content": "Please introduce tensor parallelism."}
173
+ ]
174
+ }'
175
+ ```
176
+
177
+ ## Resources
178
+
179
+ - [Step 5 Preview (FP8)](https://huggingface.co/stepfun-ai/Step-5-Preview) — FP8 release checkpoint, SGLang and Transformers deployment, reference implementation.
180
+ - [Step 5 Preview (BF16)](https://huggingface.co/stepfun-ai/Step-5-Preview-BF16) — BF16 checkpoint.
181
+ - [Step 5 Preview Model Page](https://www.stepfun.com/step-5-preview) — Model overview, evaluation context, and long-horizon case studies.
182
+ - [Documentation](https://platform.stepfun.ai/docs/en/guides/models/step-5-preview) — Official documentation and developer guides.
183
+ - [API Platform](https://platform.stepfun.ai/) — Model access and platform services.
184
+ - [StepFun Website](https://www.stepfun.com/) — Company information and model announcements.
SHA256SUMS ADDED
@@ -0,0 +1,102 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 9d459a466e9648721ef6dda4cd168d921c2b6b4e7c0a82e032d261ea9854ef2e chat_template.jinja
2
+ a20799ec92a0dec2a425dc6c64d28c280a945f37166bb331d30c2a1b4a448b0e config.json
3
+ f56bf6dd15b5b5675ff75bcc7dd271ef3353ad3ff0f68360e71389e3f5d03a8b hf_quant_config.json
4
+ 5798d715bcae0a5cdd85e963ae0eb4f1c5bbbe4704fc24423c4b941ca12f8e1b model-extra-000.safetensors
5
+ 3d8965d16468a015de1d587a660e6b44d202b1253f1141535184e27e2f554c37 model-extra-001.safetensors
6
+ dce1dfebe142c11ecc00aee3632a7573f2188b13b5b9ff87057f8042055b3a69 model-layer-000.safetensors
7
+ 1e640e5a6d780325faf9469a21ff7e0cc0bc3370f237e56b4cb56c305818b7c1 model-layer-001.safetensors
8
+ e0d41a65430b605422d082f9d1e87fe28bf206cd16019de17b9c253c7af47abb model-layer-002.safetensors
9
+ 2f353dfce8ee09c3ee7b8f91e3f506404b086219227949b4f53c2196572192b9 model-layer-003.safetensors
10
+ d42d0a4a6dea4c3c38e9474d520a8917f865a3c0eae1008c01525850a89a217a model-layer-004.safetensors
11
+ 60d8123ce6a64a4bba3ca1685aee1e4ae3e23ef0553100737373dc5ab306b2a9 model-layer-005.safetensors
12
+ 643cf15d072d1bac1ff9e4710fc07b916a96a216cd83f43dbd70795e0a350f82 model-layer-006.safetensors
13
+ 6b2d23afc429640dcbb2bc3e3ef59aab024b428f4462c4b9743e121d4a7e6332 model-layer-007.safetensors
14
+ 99e77cfc79e3ea2773a2c9bd756a15ff024fe6c23ff4a36820aa8ae24a4e599c model-layer-008.safetensors
15
+ 7a4d84b5062ca545e8128bb0823b2608320d54e70c6c1bc4c736b2e6db2a1f3e model-layer-009.safetensors
16
+ 3095ede9e5aeb62e68e518909d3b751b7dd493071535b4653d27b7c96b1f2658 model-layer-010.safetensors
17
+ 102b9b9a149d9fa55de5a5949a0c3593062e629d0014bca0f04c8417eb10a4e5 model-layer-011.safetensors
18
+ 406bf111c50354d2b15442b35cdbce40d21a5cb0ccbc9a9a8a594a7a9c0585a4 model-layer-012.safetensors
19
+ e1c89907c1f6954abbdd73044d98a790235816f5aafe941792156ee330ac8239 model-layer-013.safetensors
20
+ 38721f4f83a59c18b341e8c4ff6d90ae9919362775684411fda2b97161f221f4 model-layer-014.safetensors
21
+ 11d012e3fd33540001a953297e65893e51871ec57c4841eaf718ace795080d1b model-layer-015.safetensors
22
+ d78acb313810ec483a52e717518940a61e81bd1422c9069c26f0fe07c47b1f2e model-layer-016.safetensors
23
+ 303d7df94312a04254ab5c1c99567f08ac73036bc2c8555b9931146ce4e3bd95 model-layer-017.safetensors
24
+ 5bcef3cf91af47d01cccfbf66a4ebda584fda1c7a1220760a2cb9997181aaf94 model-layer-018.safetensors
25
+ 48846cd2449c553fd36d6356f88a98106f44455595db5908cf3fa3a972cf3ce5 model-layer-019.safetensors
26
+ 2af7fc35dbd4ca516cbf5a4bbdd22776ad17fcfffe3dae33667e136a7c279d5a model-layer-020.safetensors
27
+ 69edfddb54ae974e8b441afab64dc206e0f2815b99b946bc33463a84abf1fee4 model-layer-021.safetensors
28
+ 8f699ac13783e616ccd5796ce730ef5e7f4523ad6c1e26b4843135ef547efc0f model-layer-022.safetensors
29
+ dfd3dd480637aec3ace46b38d95f1b88b95d056e6d30a3dde47f9a328e0a1d33 model-layer-023.safetensors
30
+ 86067c89c66fe0c30582845d894b74fa4f7c193d039b931e79b3a120ea657294 model-layer-024.safetensors
31
+ c9140bf3220418051f0d2902837972b897daa60c091b4ee9fa805cd256b580e0 model-layer-025.safetensors
32
+ 08870fb96c5e55cdf4dac64d1caa80f2e6a4938302b9dcb24670f5436f88f1cf model-layer-026.safetensors
33
+ 7a22518fb3c7e931db94e65d5448f170d57f80cc3d997a3fd5a81534fc1dfdce model-layer-027.safetensors
34
+ 9a4ebe2e48ae6ecc3ee3cfbfdacfd1c8710684f51d79c0dd71c0604649be2852 model-layer-028.safetensors
35
+ f63559c9dc6c7c855c10b7b7fab564d1725ba173b38b0cf8f6561a80247f6528 model-layer-029.safetensors
36
+ c88b930d6a6414376e67908e8f8e9d2c37dc9338311e075122205114667b1bef model-layer-030.safetensors
37
+ 665f470193fceb71c67b536c9191697b9a35bb2c775104ac8e22cee187798b85 model-layer-031.safetensors
38
+ c69adc8e9e63dd5db2d6e0dba8aa268d531c5479945c37957ca82a8211edb344 model-layer-032.safetensors
39
+ b4f411dd89049a82efdc5ffe9ce9fbe8533a7659e586ec35dc2329a39d6fcf99 model-layer-033.safetensors
40
+ a1b8d02f96d21ee997b41a6cafd99d5c30bce9f6fba74832dc91c17ef7ba61ac model-layer-034.safetensors
41
+ fc9ec1882f7a2a9ddefc28431d6fdd433d8eed08df7a1edd8f0ca36bd5e758ca model-layer-035.safetensors
42
+ 3f32a245c0801f6b823b90af9024457c6d2f1eead22c31c9c6f082b618538e58 model-layer-036.safetensors
43
+ 59c173a7321479b6b9f876b0798e8bb7b4305b942347b30d039b9f5111aa9687 model-layer-037.safetensors
44
+ 13bf83586b57409d55a59627551970df202616537cc350d6ce31ff2c81050b0a model-layer-038.safetensors
45
+ c95cc0c2f90c2ca88b1ec1e135ddf803396c20698d283006bfd2154c4332a7f6 model-layer-039.safetensors
46
+ 6cda0b2c873f70465cc65ac6c21911eeb86a931e09bb6bb5c38ce057d99e6641 model-layer-040.safetensors
47
+ 0cbdd107104b0e8e8c4869eac8f50a17bdf658673bb731666c5c8f7c1ca1f8ac model-layer-041.safetensors
48
+ 2f76095d8c25b37acebb969879bca54c4856738b6b80efd6b7df05f55b1b4f9e model-layer-042.safetensors
49
+ b4a6c1197603472f66bdb6e333443a23b60ee9c0d82d60368b134b8d52ba6c29 model-layer-043.safetensors
50
+ 5aa83726f4dccbff16206ea757ebd15637752c386e0929f40279748033152ff5 model-layer-044.safetensors
51
+ 4536ded306c6468662c2533385492df5c4941c617d5c3b46110b04172a10051e model-layer-045.safetensors
52
+ 559986f9ae5437c423b9747b37f2cf893b1aa9a7138704341599a42f93ec54d7 model-layer-046.safetensors
53
+ 581886db88f80ef6571962815e0c094e2c980666462415d8820f72692c79aa16 model-layer-047.safetensors
54
+ f94d46592a9ac50d99b7ff8f2eda5a211c9b1ffd0907eabc5c5474e84eb29753 model-layer-048.safetensors
55
+ 03af787ce3eac23924fe0efc8ba16eec0c3c33174cdcbb16706d4124b3318a4d model-layer-049.safetensors
56
+ 12227d354de4ea59cdf2179df7a393c2d6a88a279b576b5b7af8474fd88f5996 model-layer-050.safetensors
57
+ 0fa4ecd33b4375413751f44d9d6ac1a1db200957975cc0190800e7c87598e217 model-layer-051.safetensors
58
+ fc445dad410f2a87eb56362c2e93833e260737fbf3d27916f4bfbdf1fe0e3dd9 model-layer-052.safetensors
59
+ d31c78e255c311b98d1434d2b4c2e8313c7cf615ae83dd11657c6c460f341b38 model-layer-053.safetensors
60
+ 0741375e87281b507a21535cc8a5e5f51e96eacf1c23643e64d7d0ef2792ba26 model-layer-054.safetensors
61
+ 3881bd40be42cf77d6f17affae4704e086cb67e914d4a66c27ac6af52f880979 model-layer-055.safetensors
62
+ cb9751c2e205038d9b23c9a2e78847db6cd61f7252992bf3b96129ab06d6e048 model-layer-056.safetensors
63
+ ef83d304f9f9d4b8ccdbc9e5260a74e15ca41fb37d8df983f888f6e84e07be23 model-layer-057.safetensors
64
+ 538d1b1b1c5ee7dbd999e62cab679a7d74c3878a96f468db692e97c1a417119d model-layer-058.safetensors
65
+ e169165f09aadeb684dc61578fe01fc6d630599590f89195b31d52c4b314cc89 model-layer-059.safetensors
66
+ f9065387fc1e3952815dd1f388779dfb794de3b367df255173a9bcb4d259d0aa model-layer-060.safetensors
67
+ ec14ad28742957af160156325212f69f924c080427988da4de930e7a0877308b model-layer-061.safetensors
68
+ 1a6c7fa478f59baaaf05b7efc40d6b260ae5c6283dacaf0b1adca2008add38a4 model-layer-062.safetensors
69
+ 6f8e12b3e2589d404bc0320fcd7a9deeb3234f6b7fe86ce7afeb864b41732fb2 model-layer-063.safetensors
70
+ 3f37b50de2837fd96ee9d6642c4cd3eb876376472d23b28ec36e66a24f0b6037 model-layer-064.safetensors
71
+ f17b47d509e4f70cdd9424670e0c48aae66cbfd76d9ebcfc392d9820024ae8af model-layer-065.safetensors
72
+ de2dc2a31af486f39fbb6e2116ea8e6a482ba39611c0b7b298d9e679140bcfea model-layer-066.safetensors
73
+ 5e0a0bf6df9d2bed5358b065456087d18078563b1d58673686568da3f65a9ad0 model-layer-067.safetensors
74
+ 5bde6af3dec2dc4ca394a3da4f3668268578d09fc2822e642b0e165ebb06d6e1 model-layer-068.safetensors
75
+ 64db0b6de77bb099c0fe49f13b03365e0dfd8c43eb6e06996550c02970fc9f02 model-layer-069.safetensors
76
+ 8f3307d3dccef1f131516651f0166da78e1bf5dd8d8677cd78defe57c62207b5 model-layer-070.safetensors
77
+ a3f08feb95ecf2523b994be9e80c1d67d498205dbe70fa035e72657aa7765cc3 model-layer-071.safetensors
78
+ 53603bfaf58d55174304acceca4ee39a0daa04511ea7d1a518e51c71c69fd56a model-layer-072.safetensors
79
+ d27307d95eab211862ac1b2c74936578436ee6bd8f8ed472bdd0fae1adac226d model-layer-073.safetensors
80
+ 752cc3e9c037fad6d0ca73620a2174bd54124e37f903c5f56875db673978fcb1 model-layer-074.safetensors
81
+ 18a33ef61645af1c961fee3821e911ef09c873af242052e2ad851a2d254806e5 model-layer-075.safetensors
82
+ af1b14510d8d60e6387d2fbdd4a4b0f566c71ab171cff33f2ee76b50fc59bb98 model-layer-076.safetensors
83
+ eee4f057b9cd894cbc5c22736d23e62e7b457122dd3ecf5f5de00256412b759a model-layer-077.safetensors
84
+ 89f762928ad68864c458a73c19074a5d13bd5b9262a22b844bb4f3f973fef72e model-layer-078.safetensors
85
+ e3eee35b9b322f323026c716889b2935bd31f6314bdcf121b4867c84917b6a40 model-layer-079.safetensors
86
+ ba1f77590d19c84467b05ea54b7076b9ff5f352964020fbfc7a498c6ccbf3f36 model-layer-080.safetensors
87
+ 8c8eec612fde8a89c9ca007fec6cbfca722d5aeec3f87ec9d184e329f6581e3b model-layer-081.safetensors
88
+ 768265cede3eb1a96b686e4647a32ce149e454e0bbd4da5ce46ba419e8d4d07f model-layer-082.safetensors
89
+ 04fdf3647932cd412a4bfbad3f9e1dc80f1b548e97ab4af05675c7f0a284d89b model-layer-083.safetensors
90
+ 4b7aa17c4ad72fb6a39043e9ba461c047f1dc757740101da27b3466786ad0889 model-layer-084.safetensors
91
+ 6795f2546f3fe6272d5d5fda80cae6e075178404db261c39754ab4a152adfb73 model-layer-085.safetensors
92
+ 2f205d7d1cc29f39263b92ef936a68567d768b52e94b3be886d121a2ee5efc88 model-layer-086.safetensors
93
+ cf8b61dc8709acf36a81d906b64cd9f062f3b6e07c18e48bff96a79aa88de177 model-layer-087.safetensors
94
+ 24c0cef80ba07fff6bf321fd3994e74e11df6728a073aaa019c10c717f8b29f6 model-layer-088.safetensors
95
+ ae98298dbf92a7129c8fc07d7fbe5f2e779d63f9a09cedf80ac15b7e71084f1c model-layer-089.safetensors
96
+ 178ae778101f5dbc87e7467ca861fc1ac0b1b19c26a20bb88afd9d8edc5eb819 model-layer-090.safetensors
97
+ 1cfb4c3077c3f155c516cbf89ecc5ac03daf0a725beb929c9e69d9444bdf96df model-layer-091.safetensors
98
+ 3bb6e44845d1cc16e56c0eeae2eed781067215083f2d21c504fbece0c1cec8ea model-vit-000.safetensors
99
+ bec6dd828b32297ea75fa6a28c23c2fb9e125f542c0c865b7c852739894e1053 model.safetensors.index.json
100
+ d47424bda11df4cedc3f9458915c465a28e601d3b7df0e78f6dff4d7727006c4 special_tokens_map.json
101
+ b564c620eb77fa11d0926011c2202347d6cfc358d79724ee04ae7007e13636f0 tokenizer.json
102
+ 78202af487f4d4360e8d15cb0506d60718f8599770599c6f28f7f1fa045a591f tokenizer_config.json
chat_template.jinja ADDED
@@ -0,0 +1,97 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {% macro render_message_content(message) %}{% if message.content is none %}{{- '' }}{% elif message.content is string %}{{- message.content }}{% elif message.content is mapping %}{{- message.content['value'] if 'value' in message.content else message.content['text'] }}{% elif message.content is iterable %}{% set ns = namespace(needs_text_separator=false) %}{% for item in message.content %}{% if item.type == 'text' %}{% if ns.needs_text_separator %}{{- ' ' }}{% endif %}{{- item['value'] if 'value' in item else item['text'] }}{% set ns.needs_text_separator = true %}{% elif item.type == 'image' %}<im_patch>{% set ns.needs_text_separator = false %}{% endif %}{% endfor %}{% endif %}{% endmacro %}
2
+ {{bos_token}}{%- if tools %}
3
+ {{- '<|im_start|>system\n' }}
4
+ {%- if reasoning_effort is defined %}
5
+ {{- "Reasoning: " + reasoning_effort + '\n\n' }}
6
+ {%- endif %}
7
+ {%- if messages[0].role == 'system' %}
8
+ {{- render_message_content(messages[0]) + '\n\n' }}
9
+ {%- endif %}
10
+ {{- "# Tools\n\nYou have access to the following functions in JSONSchema format:\n\n<tools>" }}
11
+ {%- for tool in tools %}
12
+ {{- "\n" }}
13
+ {{- tool | tojson(ensure_ascii=False) }}
14
+ {%- endfor %}
15
+ {{- "\n</tools>\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...>\n...\n</function> block must be nested within <tool_call>\n...\n</tool_call> XML tags\n- Required parameters MUST be specified\n</IMPORTANT><|im_end|>\n" }}
16
+ {%- else %}
17
+ {%- if messages[0].role == 'system' %}
18
+ {{- '<|im_start|>system\n' }}
19
+ {%- if reasoning_effort is defined %}
20
+ {{- "Reasoning: " + reasoning_effort + '\n\n' }}
21
+ {%- endif %}
22
+ {{- render_message_content(messages[0]) + '<|im_end|>\n' }}
23
+ {%- elif reasoning_effort is defined %}
24
+ {{- '<|im_start|>system\n' + "Reasoning: " + reasoning_effort + '\n\n' + '<|im_end|>\n' }}
25
+ {%- endif %}
26
+ {%- endif %}
27
+ {%- set preserve_thinking = preserve_thinking | default(true) %}
28
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
29
+ {%- for message in messages[::-1] %}
30
+ {%- set index = (messages|length - 1) - loop.index0 %}
31
+ {%- if ns.multi_step_tool and message.role == "user" and render_message_content(message) is string and not(render_message_content(message).startswith('<tool_response>') and render_message_content(message).endswith('</tool_response>')) %}
32
+ {%- set ns.multi_step_tool = false %}
33
+ {%- set ns.last_query_index = index %}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- for message in messages %}
37
+ {%- set content = render_message_content(message) %}
38
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
39
+ {%- set role_name = 'observation' if (message.role == "system" and not loop.first and message.name == 'observation') else message.role %}
40
+ {{- '<|im_start|>' + role_name + '\n' + content + '<|im_end|>' + '\n' }}
41
+ {%- elif message.role == "assistant" %}
42
+ {%- if message.reasoning_content is string %}
43
+ {%- set reasoning_content = message.reasoning_content %}
44
+ {%- else %}
45
+ {%- if '</think>' in content %}
46
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
47
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
48
+ {%- else %}
49
+ {%- set reasoning_content = '' %}
50
+ {%- endif %}
51
+ {%- endif %}
52
+ {%- if preserve_thinking or loop.index0 > ns.last_query_index %}
53
+ {{- '<|im_start|>' + message.role + '\n<think>' }}
54
+ {%- else %}
55
+ {{- '<|im_start|>' + message.role + '\n' }}
56
+ {%- endif %}
57
+ {%- generation %}
58
+ {%- if preserve_thinking or loop.index0 > ns.last_query_index %}
59
+ {{- reasoning_content + '</think>' + content }}
60
+ {%- else %}
61
+ {{- content }}
62
+ {%- endif %}
63
+ {%- if message.tool_calls %}
64
+ {%- for tool_call in message.tool_calls %}
65
+ {%- if tool_call.function is defined %}
66
+ {%- set tool_call = tool_call.function %}
67
+ {%- endif %}
68
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
69
+ {%- if tool_call.arguments is defined %}
70
+ {%- set arguments = tool_call.arguments | fromjson if tool_call.arguments is string else tool_call.arguments %}
71
+ {%- for args_name, args_value in arguments|items %}
72
+ {{- '<parameter=' + args_name + '>\n' }}
73
+ {%- set args_value = args_value | tojson(ensure_ascii=False) | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
74
+ {{- args_value }}
75
+ {{- '\n</parameter>\n' }}
76
+ {%- endfor %}
77
+ {%- endif %}
78
+ {{- '</function>\n</tool_call>' }}
79
+ {%- endfor %}
80
+ {%- endif %}
81
+ {{- '<|im_end|>\n' }}
82
+ {%- endgeneration %}
83
+ {%- elif message.role == "tool" %}
84
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
85
+ {{- '<|im_start|>tool_response\n' }}
86
+ {%- endif %}
87
+ {{- '<tool_response>' }}
88
+ {{- content }}
89
+ {{- '</tool_response>' }}
90
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
91
+ {{- '<|im_end|>\n' }}
92
+ {%- endif %}
93
+ {%- endif %}
94
+ {%- endfor %}
95
+ {%- if add_generation_prompt %}
96
+ {{- '<|im_start|>assistant\n<think>' }}
97
+ {%- endif %}
config.json ADDED
@@ -0,0 +1,1070 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Step5ForConditionalGeneration"
4
+ ],
5
+ "model_type": "step5_vl",
6
+ "im_end_token": "<im_end>",
7
+ "im_patch_token": "<im_patch>",
8
+ "im_start_token": "<im_start>",
9
+ "image_token_len": 169,
10
+ "patch_token_len": 81,
11
+ "image_token_id": 128001,
12
+ "understand_projector_stride": 2,
13
+ "use_im_start_end": "true",
14
+ "vision_select_layer": -1,
15
+ "projector_bias": false,
16
+ "vision_config": {
17
+ "model_type": "perception_encoder",
18
+ "image_size": 728,
19
+ "patch_size": 14,
20
+ "width": 1536,
21
+ "layers": 47,
22
+ "heads": 16,
23
+ "pool_type": "none",
24
+ "output_dim": null,
25
+ "use_cls_token": false,
26
+ "ls_init_value": 0.1,
27
+ "use_ln_post": false,
28
+ "hidden_act": "quick_gelu"
29
+ },
30
+ "text_config": {
31
+ "architectures": [
32
+ "Step5ForCausalLM"
33
+ ],
34
+ "model_type": "step5",
35
+ "att_impl_type": "GQA",
36
+ "torch_dtype": "bfloat16",
37
+ "norm_dtype": "float32",
38
+ "bos_token_id": 0,
39
+ "eos_token_id": [
40
+ 1,
41
+ 2
42
+ ],
43
+ "hidden_size": 4096,
44
+ "intermediate_size": 13824,
45
+ "num_hidden_layers": 92,
46
+ "vocab_size": 128896,
47
+ "rms_norm_eps": 1e-05,
48
+ "fp32_residual_connection": true,
49
+ "num_attention_heads": 64,
50
+ "num_attention_groups": 4,
51
+ "num_sliding_attention_heads": 64,
52
+ "head_dim": 192,
53
+ "head_dim_full_attention": 192,
54
+ "head_dim_sliding_attention": 192,
55
+ "num_attention_groups_full_attention": 4,
56
+ "num_attention_groups_sliding_attention": 4,
57
+ "rope_theta": [
58
+ 10000.0,
59
+ 10000.0,
60
+ 10000.0,
61
+ 10000000.0,
62
+ 10000.0,
63
+ 10000.0,
64
+ 10000.0,
65
+ 10000000.0,
66
+ 10000.0,
67
+ 10000.0,
68
+ 10000.0,
69
+ 10000000.0,
70
+ 10000.0,
71
+ 10000.0,
72
+ 10000.0,
73
+ 10000000.0,
74
+ 10000.0,
75
+ 10000.0,
76
+ 10000.0,
77
+ 10000000.0,
78
+ 10000.0,
79
+ 10000.0,
80
+ 10000.0,
81
+ 10000000.0,
82
+ 10000.0,
83
+ 10000.0,
84
+ 10000.0,
85
+ 10000000.0,
86
+ 10000.0,
87
+ 10000.0,
88
+ 10000.0,
89
+ 10000000.0,
90
+ 10000.0,
91
+ 10000.0,
92
+ 10000.0,
93
+ 10000000.0,
94
+ 10000.0,
95
+ 10000.0,
96
+ 10000.0,
97
+ 10000000.0,
98
+ 10000.0,
99
+ 10000.0,
100
+ 10000.0,
101
+ 10000000.0,
102
+ 10000.0,
103
+ 10000.0,
104
+ 10000.0,
105
+ 10000000.0,
106
+ 10000.0,
107
+ 10000.0,
108
+ 10000.0,
109
+ 10000000.0,
110
+ 10000.0,
111
+ 10000.0,
112
+ 10000.0,
113
+ 10000000.0,
114
+ 10000.0,
115
+ 10000.0,
116
+ 10000.0,
117
+ 10000000.0,
118
+ 10000.0,
119
+ 10000.0,
120
+ 10000.0,
121
+ 10000000.0,
122
+ 10000.0,
123
+ 10000.0,
124
+ 10000.0,
125
+ 10000000.0,
126
+ 10000.0,
127
+ 10000.0,
128
+ 10000.0,
129
+ 10000000.0,
130
+ 10000.0,
131
+ 10000.0,
132
+ 10000.0,
133
+ 10000000.0,
134
+ 10000.0,
135
+ 10000.0,
136
+ 10000.0,
137
+ 10000000.0,
138
+ 10000.0,
139
+ 10000.0,
140
+ 10000.0,
141
+ 10000000.0,
142
+ 10000.0,
143
+ 10000.0,
144
+ 10000.0,
145
+ 10000000.0,
146
+ 10000.0,
147
+ 10000.0,
148
+ 10000.0,
149
+ 10000000.0,
150
+ 10000.0,
151
+ 10000.0,
152
+ 10000.0
153
+ ],
154
+ "layer_types": [
155
+ "sliding_attention",
156
+ "sliding_attention",
157
+ "sliding_attention",
158
+ "full_attention",
159
+ "sliding_attention",
160
+ "sliding_attention",
161
+ "sliding_attention",
162
+ "full_attention",
163
+ "sliding_attention",
164
+ "sliding_attention",
165
+ "sliding_attention",
166
+ "full_attention",
167
+ "sliding_attention",
168
+ "sliding_attention",
169
+ "sliding_attention",
170
+ "full_attention",
171
+ "sliding_attention",
172
+ "sliding_attention",
173
+ "sliding_attention",
174
+ "full_attention",
175
+ "sliding_attention",
176
+ "sliding_attention",
177
+ "sliding_attention",
178
+ "full_attention",
179
+ "sliding_attention",
180
+ "sliding_attention",
181
+ "sliding_attention",
182
+ "full_attention",
183
+ "sliding_attention",
184
+ "sliding_attention",
185
+ "sliding_attention",
186
+ "full_attention",
187
+ "sliding_attention",
188
+ "sliding_attention",
189
+ "sliding_attention",
190
+ "full_attention",
191
+ "sliding_attention",
192
+ "sliding_attention",
193
+ "sliding_attention",
194
+ "full_attention",
195
+ "sliding_attention",
196
+ "sliding_attention",
197
+ "sliding_attention",
198
+ "full_attention",
199
+ "sliding_attention",
200
+ "sliding_attention",
201
+ "sliding_attention",
202
+ "full_attention",
203
+ "sliding_attention",
204
+ "sliding_attention",
205
+ "sliding_attention",
206
+ "full_attention",
207
+ "sliding_attention",
208
+ "sliding_attention",
209
+ "sliding_attention",
210
+ "full_attention",
211
+ "sliding_attention",
212
+ "sliding_attention",
213
+ "sliding_attention",
214
+ "full_attention",
215
+ "sliding_attention",
216
+ "sliding_attention",
217
+ "sliding_attention",
218
+ "full_attention",
219
+ "sliding_attention",
220
+ "sliding_attention",
221
+ "sliding_attention",
222
+ "full_attention",
223
+ "sliding_attention",
224
+ "sliding_attention",
225
+ "sliding_attention",
226
+ "full_attention",
227
+ "sliding_attention",
228
+ "sliding_attention",
229
+ "sliding_attention",
230
+ "full_attention",
231
+ "sliding_attention",
232
+ "sliding_attention",
233
+ "sliding_attention",
234
+ "full_attention",
235
+ "sliding_attention",
236
+ "sliding_attention",
237
+ "sliding_attention",
238
+ "full_attention",
239
+ "sliding_attention",
240
+ "sliding_attention",
241
+ "sliding_attention",
242
+ "full_attention",
243
+ "sliding_attention",
244
+ "sliding_attention",
245
+ "sliding_attention",
246
+ "full_attention",
247
+ "sliding_attention",
248
+ "sliding_attention",
249
+ "sliding_attention"
250
+ ],
251
+ "partial_rotary_factors": [
252
+ 1.0,
253
+ 1.0,
254
+ 1.0,
255
+ 0.3333333333333333,
256
+ 1.0,
257
+ 1.0,
258
+ 1.0,
259
+ 0.3333333333333333,
260
+ 1.0,
261
+ 1.0,
262
+ 1.0,
263
+ 0.3333333333333333,
264
+ 1.0,
265
+ 1.0,
266
+ 1.0,
267
+ 0.3333333333333333,
268
+ 1.0,
269
+ 1.0,
270
+ 1.0,
271
+ 0.3333333333333333,
272
+ 1.0,
273
+ 1.0,
274
+ 1.0,
275
+ 0.3333333333333333,
276
+ 1.0,
277
+ 1.0,
278
+ 1.0,
279
+ 0.3333333333333333,
280
+ 1.0,
281
+ 1.0,
282
+ 1.0,
283
+ 0.3333333333333333,
284
+ 1.0,
285
+ 1.0,
286
+ 1.0,
287
+ 0.3333333333333333,
288
+ 1.0,
289
+ 1.0,
290
+ 1.0,
291
+ 0.3333333333333333,
292
+ 1.0,
293
+ 1.0,
294
+ 1.0,
295
+ 0.3333333333333333,
296
+ 1.0,
297
+ 1.0,
298
+ 1.0,
299
+ 0.3333333333333333,
300
+ 1.0,
301
+ 1.0,
302
+ 1.0,
303
+ 0.3333333333333333,
304
+ 1.0,
305
+ 1.0,
306
+ 1.0,
307
+ 0.3333333333333333,
308
+ 1.0,
309
+ 1.0,
310
+ 1.0,
311
+ 0.3333333333333333,
312
+ 1.0,
313
+ 1.0,
314
+ 1.0,
315
+ 0.3333333333333333,
316
+ 1.0,
317
+ 1.0,
318
+ 1.0,
319
+ 0.3333333333333333,
320
+ 1.0,
321
+ 1.0,
322
+ 1.0,
323
+ 0.3333333333333333,
324
+ 1.0,
325
+ 1.0,
326
+ 1.0,
327
+ 0.3333333333333333,
328
+ 1.0,
329
+ 1.0,
330
+ 1.0,
331
+ 0.3333333333333333,
332
+ 1.0,
333
+ 1.0,
334
+ 1.0,
335
+ 0.3333333333333333,
336
+ 1.0,
337
+ 1.0,
338
+ 1.0,
339
+ 0.3333333333333333,
340
+ 1.0,
341
+ 1.0,
342
+ 1.0,
343
+ 0.3333333333333333,
344
+ 1.0,
345
+ 1.0,
346
+ 1.0
347
+ ],
348
+ "sliding_window": 512,
349
+ "attention_other_setting": {
350
+ "attention_type": "sliding_attention",
351
+ "num_attention_heads": 64,
352
+ "num_attention_groups": 4,
353
+ "head_dim": 192,
354
+ "true_head_dim": 192
355
+ },
356
+ "use_head_wise_attn_gate": true,
357
+ "num_nextn_predict_layers": 3,
358
+ "use_moe": true,
359
+ "moe_intermediate_size": 1536,
360
+ "share_expert_dim": 1536,
361
+ "moe_num_experts": 352,
362
+ "moe_top_k": 8,
363
+ "num_experts_per_tok": 8,
364
+ "moe_layer_list": [
365
+ 3,
366
+ 4,
367
+ 5,
368
+ 6,
369
+ 7,
370
+ 8,
371
+ 9,
372
+ 10,
373
+ 11,
374
+ 12,
375
+ 13,
376
+ 14,
377
+ 15,
378
+ 16,
379
+ 17,
380
+ 18,
381
+ 19,
382
+ 20,
383
+ 21,
384
+ 22,
385
+ 23,
386
+ 24,
387
+ 25,
388
+ 26,
389
+ 27,
390
+ 28,
391
+ 29,
392
+ 30,
393
+ 31,
394
+ 32,
395
+ 33,
396
+ 34,
397
+ 35,
398
+ 36,
399
+ 37,
400
+ 38,
401
+ 39,
402
+ 40,
403
+ 41,
404
+ 42,
405
+ 43,
406
+ 44,
407
+ 45,
408
+ 46,
409
+ 47,
410
+ 48,
411
+ 49,
412
+ 50,
413
+ 51,
414
+ 52,
415
+ 53,
416
+ 54,
417
+ 55,
418
+ 56,
419
+ 57,
420
+ 58,
421
+ 59,
422
+ 60,
423
+ 61,
424
+ 62,
425
+ 63,
426
+ 64,
427
+ 65,
428
+ 66,
429
+ 67,
430
+ 68,
431
+ 69,
432
+ 70,
433
+ 71,
434
+ 72,
435
+ 73,
436
+ 74,
437
+ 75,
438
+ 76,
439
+ 77,
440
+ 78,
441
+ 79,
442
+ 80,
443
+ 81,
444
+ 82,
445
+ 83,
446
+ 84,
447
+ 85,
448
+ 86,
449
+ 87,
450
+ 88,
451
+ 89,
452
+ 90
453
+ ],
454
+ "moe_layers_enum": "3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90",
455
+ "use_moe_router_bias": true,
456
+ "need_fp32_gate": true,
457
+ "swiglu_limits": [
458
+ 7.0,
459
+ 7.0,
460
+ 7.0,
461
+ 7.0,
462
+ 7.0,
463
+ 7.0,
464
+ 7.0,
465
+ 7.0,
466
+ 7.0,
467
+ 7.0,
468
+ 7.0,
469
+ 7.0,
470
+ 7.0,
471
+ 7.0,
472
+ 7.0,
473
+ 7.0,
474
+ 7.0,
475
+ 7.0,
476
+ 7.0,
477
+ 7.0,
478
+ 7.0,
479
+ 7.0,
480
+ 7.0,
481
+ 7.0,
482
+ 7.0,
483
+ 7.0,
484
+ 7.0,
485
+ 7.0,
486
+ 7.0,
487
+ 7.0,
488
+ 7.0,
489
+ 7.0,
490
+ 7.0,
491
+ 7.0,
492
+ 7.0,
493
+ 7.0,
494
+ 7.0,
495
+ 7.0,
496
+ 7.0,
497
+ 7.0,
498
+ 7.0,
499
+ 7.0,
500
+ 7.0,
501
+ 7.0,
502
+ 7.0,
503
+ 7.0,
504
+ 7.0,
505
+ 7.0,
506
+ 7.0,
507
+ 7.0,
508
+ 7.0,
509
+ 7.0,
510
+ 7.0,
511
+ 7.0,
512
+ 7.0,
513
+ 7.0,
514
+ 7.0,
515
+ 7.0,
516
+ 7.0,
517
+ 7.0,
518
+ 7.0,
519
+ 7.0,
520
+ 7.0,
521
+ 7.0,
522
+ 7.0,
523
+ 7.0,
524
+ 7.0,
525
+ 7.0,
526
+ 7.0,
527
+ 7.0,
528
+ 7.0,
529
+ 7.0,
530
+ 7.0,
531
+ 7.0,
532
+ 7.0,
533
+ 7.0,
534
+ 7.0,
535
+ 7.0,
536
+ 7.0,
537
+ 7.0,
538
+ 7.0,
539
+ 7.0,
540
+ 7.0,
541
+ 7.0,
542
+ 7.0,
543
+ 7.0,
544
+ 7.0,
545
+ 7.0,
546
+ 7.0,
547
+ 7.0,
548
+ 7.0,
549
+ 7.0,
550
+ 7.0,
551
+ 7.0,
552
+ 7.0
553
+ ],
554
+ "swiglu_limits_shared": [
555
+ 7.0,
556
+ 7.0,
557
+ 7.0,
558
+ 7.0,
559
+ 7.0,
560
+ 7.0,
561
+ 7.0,
562
+ 7.0,
563
+ 7.0,
564
+ 7.0,
565
+ 7.0,
566
+ 7.0,
567
+ 7.0,
568
+ 7.0,
569
+ 7.0,
570
+ 7.0,
571
+ 7.0,
572
+ 7.0,
573
+ 7.0,
574
+ 7.0,
575
+ 7.0,
576
+ 7.0,
577
+ 7.0,
578
+ 7.0,
579
+ 7.0,
580
+ 7.0,
581
+ 7.0,
582
+ 7.0,
583
+ 7.0,
584
+ 7.0,
585
+ 7.0,
586
+ 7.0,
587
+ 7.0,
588
+ 7.0,
589
+ 7.0,
590
+ 7.0,
591
+ 7.0,
592
+ 7.0,
593
+ 7.0,
594
+ 7.0,
595
+ 7.0,
596
+ 7.0,
597
+ 7.0,
598
+ 7.0,
599
+ 7.0,
600
+ 7.0,
601
+ 7.0,
602
+ 7.0,
603
+ 7.0,
604
+ 7.0,
605
+ 7.0,
606
+ 7.0,
607
+ 7.0,
608
+ 7.0,
609
+ 7.0,
610
+ 7.0,
611
+ 7.0,
612
+ 7.0,
613
+ 7.0,
614
+ 7.0,
615
+ 7.0,
616
+ 7.0,
617
+ 7.0,
618
+ 7.0,
619
+ 7.0,
620
+ 7.0,
621
+ 7.0,
622
+ 7.0,
623
+ 7.0,
624
+ 7.0,
625
+ 7.0,
626
+ 7.0,
627
+ 7.0,
628
+ 7.0,
629
+ 7.0,
630
+ 7.0,
631
+ 7.0,
632
+ 7.0,
633
+ 7.0,
634
+ 7.0,
635
+ 7.0,
636
+ 7.0,
637
+ 7.0,
638
+ 7.0,
639
+ 7.0,
640
+ 7.0,
641
+ 7.0,
642
+ 7.0,
643
+ 7.0,
644
+ 7.0,
645
+ 7.0,
646
+ 7.0,
647
+ 7.0,
648
+ 7.0,
649
+ 7.0
650
+ ],
651
+ "max_seq_len": 1048576,
652
+ "max_position_embeddings": 1048576,
653
+ "norm_expert_weight": true,
654
+ "moe_router_scaling_factor": 3.0,
655
+ "rope_scaling": {
656
+ "rope_type": "llama3",
657
+ "factor": 1.0,
658
+ "original_max_position_embeddings": 1048576,
659
+ "low_freq_factor": 1.0,
660
+ "high_freq_factor": 32.0
661
+ },
662
+ "yarn_only_types": [
663
+ "full_attention"
664
+ ],
665
+ "sparse_config": {
666
+ "enabled": true,
667
+ "proxy_dim": 256,
668
+ "sparse_indexer_rope_dim": 32,
669
+ "sparse_indexer_use_rope": true,
670
+ "sparse_indexer_num_heads": 16,
671
+ "sparse_indexer_num_k_heads": 1,
672
+ "sparse_indexer_q_norm_type": "rmsnorm",
673
+ "sparse_indexer_k_norm_type": "layernorm",
674
+ "sparse_indexer_csa_z_norm_type": "none",
675
+ "sparse_indexer_softmax_variant": "ssmax",
676
+ "sparse_indexer_ssmax_s_granularity": "q_head",
677
+ "num_provider_groups": 4,
678
+ "topk": 512,
679
+ "region_block_size": 8,
680
+ "compression_method": "csa_block_compress",
681
+ "attention_impl": "sparse_gqa",
682
+ "apply_to_layer_types": [
683
+ "full_attention"
684
+ ]
685
+ }
686
+ },
687
+ "quantization_config": {
688
+ "quant_method": "modelopt",
689
+ "quant_algo": "NVFP4",
690
+ "group_size": 16,
691
+ "kv_cache_quant_algo": null,
692
+ "ignore": [
693
+ "lm_head",
694
+ "model.embed_tokens",
695
+ "model.norm",
696
+ "*layers.0.self_attn*",
697
+ "*layers.0.mlp*",
698
+ "*layers.1.self_attn*",
699
+ "*layers.1.mlp*",
700
+ "*layers.2.self_attn*",
701
+ "*layers.2.mlp*",
702
+ "*layers.3.self_attn*",
703
+ "*layers.3.share_expert*",
704
+ "*layers.3.moe.share_expert*",
705
+ "*layers.3.moe.gate*",
706
+ "*layers.4.self_attn*",
707
+ "*layers.4.share_expert*",
708
+ "*layers.4.moe.share_expert*",
709
+ "*layers.4.moe.gate*",
710
+ "*layers.5.self_attn*",
711
+ "*layers.5.share_expert*",
712
+ "*layers.5.moe.share_expert*",
713
+ "*layers.5.moe.gate*",
714
+ "*layers.6.self_attn*",
715
+ "*layers.6.share_expert*",
716
+ "*layers.6.moe.share_expert*",
717
+ "*layers.6.moe.gate*",
718
+ "*layers.7.self_attn*",
719
+ "*layers.7.share_expert*",
720
+ "*layers.7.moe.share_expert*",
721
+ "*layers.7.moe.gate*",
722
+ "*layers.8.self_attn*",
723
+ "*layers.8.share_expert*",
724
+ "*layers.8.moe.share_expert*",
725
+ "*layers.8.moe.gate*",
726
+ "*layers.9.self_attn*",
727
+ "*layers.9.share_expert*",
728
+ "*layers.9.moe.share_expert*",
729
+ "*layers.9.moe.gate*",
730
+ "*layers.10.self_attn*",
731
+ "*layers.10.share_expert*",
732
+ "*layers.10.moe.share_expert*",
733
+ "*layers.10.moe.gate*",
734
+ "*layers.11.self_attn*",
735
+ "*layers.11.share_expert*",
736
+ "*layers.11.moe.share_expert*",
737
+ "*layers.11.moe.gate*",
738
+ "*layers.12.self_attn*",
739
+ "*layers.12.share_expert*",
740
+ "*layers.12.moe.share_expert*",
741
+ "*layers.12.moe.gate*",
742
+ "*layers.13.self_attn*",
743
+ "*layers.13.share_expert*",
744
+ "*layers.13.moe.share_expert*",
745
+ "*layers.13.moe.gate*",
746
+ "*layers.14.self_attn*",
747
+ "*layers.14.share_expert*",
748
+ "*layers.14.moe.share_expert*",
749
+ "*layers.14.moe.gate*",
750
+ "*layers.15.self_attn*",
751
+ "*layers.15.share_expert*",
752
+ "*layers.15.moe.share_expert*",
753
+ "*layers.15.moe.gate*",
754
+ "*layers.16.self_attn*",
755
+ "*layers.16.share_expert*",
756
+ "*layers.16.moe.share_expert*",
757
+ "*layers.16.moe.gate*",
758
+ "*layers.17.self_attn*",
759
+ "*layers.17.share_expert*",
760
+ "*layers.17.moe.share_expert*",
761
+ "*layers.17.moe.gate*",
762
+ "*layers.18.self_attn*",
763
+ "*layers.18.share_expert*",
764
+ "*layers.18.moe.share_expert*",
765
+ "*layers.18.moe.gate*",
766
+ "*layers.19.self_attn*",
767
+ "*layers.19.share_expert*",
768
+ "*layers.19.moe.share_expert*",
769
+ "*layers.19.moe.gate*",
770
+ "*layers.20.self_attn*",
771
+ "*layers.20.share_expert*",
772
+ "*layers.20.moe.share_expert*",
773
+ "*layers.20.moe.gate*",
774
+ "*layers.21.self_attn*",
775
+ "*layers.21.share_expert*",
776
+ "*layers.21.moe.share_expert*",
777
+ "*layers.21.moe.gate*",
778
+ "*layers.22.self_attn*",
779
+ "*layers.22.share_expert*",
780
+ "*layers.22.moe.share_expert*",
781
+ "*layers.22.moe.gate*",
782
+ "*layers.23.self_attn*",
783
+ "*layers.23.share_expert*",
784
+ "*layers.23.moe.share_expert*",
785
+ "*layers.23.moe.gate*",
786
+ "*layers.24.self_attn*",
787
+ "*layers.24.share_expert*",
788
+ "*layers.24.moe.share_expert*",
789
+ "*layers.24.moe.gate*",
790
+ "*layers.25.self_attn*",
791
+ "*layers.25.share_expert*",
792
+ "*layers.25.moe.share_expert*",
793
+ "*layers.25.moe.gate*",
794
+ "*layers.26.self_attn*",
795
+ "*layers.26.share_expert*",
796
+ "*layers.26.moe.share_expert*",
797
+ "*layers.26.moe.gate*",
798
+ "*layers.27.self_attn*",
799
+ "*layers.27.share_expert*",
800
+ "*layers.27.moe.share_expert*",
801
+ "*layers.27.moe.gate*",
802
+ "*layers.28.self_attn*",
803
+ "*layers.28.share_expert*",
804
+ "*layers.28.moe.share_expert*",
805
+ "*layers.28.moe.gate*",
806
+ "*layers.29.self_attn*",
807
+ "*layers.29.share_expert*",
808
+ "*layers.29.moe.share_expert*",
809
+ "*layers.29.moe.gate*",
810
+ "*layers.30.self_attn*",
811
+ "*layers.30.share_expert*",
812
+ "*layers.30.moe.share_expert*",
813
+ "*layers.30.moe.gate*",
814
+ "*layers.31.self_attn*",
815
+ "*layers.31.share_expert*",
816
+ "*layers.31.moe.share_expert*",
817
+ "*layers.31.moe.gate*",
818
+ "*layers.32.self_attn*",
819
+ "*layers.32.share_expert*",
820
+ "*layers.32.moe.share_expert*",
821
+ "*layers.32.moe.gate*",
822
+ "*layers.33.self_attn*",
823
+ "*layers.33.share_expert*",
824
+ "*layers.33.moe.share_expert*",
825
+ "*layers.33.moe.gate*",
826
+ "*layers.34.self_attn*",
827
+ "*layers.34.share_expert*",
828
+ "*layers.34.moe.share_expert*",
829
+ "*layers.34.moe.gate*",
830
+ "*layers.35.self_attn*",
831
+ "*layers.35.share_expert*",
832
+ "*layers.35.moe.share_expert*",
833
+ "*layers.35.moe.gate*",
834
+ "*layers.36.self_attn*",
835
+ "*layers.36.share_expert*",
836
+ "*layers.36.moe.share_expert*",
837
+ "*layers.36.moe.gate*",
838
+ "*layers.37.self_attn*",
839
+ "*layers.37.share_expert*",
840
+ "*layers.37.moe.share_expert*",
841
+ "*layers.37.moe.gate*",
842
+ "*layers.38.self_attn*",
843
+ "*layers.38.share_expert*",
844
+ "*layers.38.moe.share_expert*",
845
+ "*layers.38.moe.gate*",
846
+ "*layers.39.self_attn*",
847
+ "*layers.39.share_expert*",
848
+ "*layers.39.moe.share_expert*",
849
+ "*layers.39.moe.gate*",
850
+ "*layers.40.self_attn*",
851
+ "*layers.40.share_expert*",
852
+ "*layers.40.moe.share_expert*",
853
+ "*layers.40.moe.gate*",
854
+ "*layers.41.self_attn*",
855
+ "*layers.41.share_expert*",
856
+ "*layers.41.moe.share_expert*",
857
+ "*layers.41.moe.gate*",
858
+ "*layers.42.self_attn*",
859
+ "*layers.42.share_expert*",
860
+ "*layers.42.moe.share_expert*",
861
+ "*layers.42.moe.gate*",
862
+ "*layers.43.self_attn*",
863
+ "*layers.43.share_expert*",
864
+ "*layers.43.moe.share_expert*",
865
+ "*layers.43.moe.gate*",
866
+ "*layers.44.self_attn*",
867
+ "*layers.44.share_expert*",
868
+ "*layers.44.moe.share_expert*",
869
+ "*layers.44.moe.gate*",
870
+ "*layers.45.self_attn*",
871
+ "*layers.45.share_expert*",
872
+ "*layers.45.moe.share_expert*",
873
+ "*layers.45.moe.gate*",
874
+ "*layers.46.self_attn*",
875
+ "*layers.46.share_expert*",
876
+ "*layers.46.moe.share_expert*",
877
+ "*layers.46.moe.gate*",
878
+ "*layers.47.self_attn*",
879
+ "*layers.47.share_expert*",
880
+ "*layers.47.moe.share_expert*",
881
+ "*layers.47.moe.gate*",
882
+ "*layers.48.self_attn*",
883
+ "*layers.48.share_expert*",
884
+ "*layers.48.moe.share_expert*",
885
+ "*layers.48.moe.gate*",
886
+ "*layers.49.self_attn*",
887
+ "*layers.49.share_expert*",
888
+ "*layers.49.moe.share_expert*",
889
+ "*layers.49.moe.gate*",
890
+ "*layers.50.self_attn*",
891
+ "*layers.50.share_expert*",
892
+ "*layers.50.moe.share_expert*",
893
+ "*layers.50.moe.gate*",
894
+ "*layers.51.self_attn*",
895
+ "*layers.51.share_expert*",
896
+ "*layers.51.moe.share_expert*",
897
+ "*layers.51.moe.gate*",
898
+ "*layers.52.self_attn*",
899
+ "*layers.52.share_expert*",
900
+ "*layers.52.moe.share_expert*",
901
+ "*layers.52.moe.gate*",
902
+ "*layers.53.self_attn*",
903
+ "*layers.53.share_expert*",
904
+ "*layers.53.moe.share_expert*",
905
+ "*layers.53.moe.gate*",
906
+ "*layers.54.self_attn*",
907
+ "*layers.54.share_expert*",
908
+ "*layers.54.moe.share_expert*",
909
+ "*layers.54.moe.gate*",
910
+ "*layers.55.self_attn*",
911
+ "*layers.55.share_expert*",
912
+ "*layers.55.moe.share_expert*",
913
+ "*layers.55.moe.gate*",
914
+ "*layers.56.self_attn*",
915
+ "*layers.56.share_expert*",
916
+ "*layers.56.moe.share_expert*",
917
+ "*layers.56.moe.gate*",
918
+ "*layers.57.self_attn*",
919
+ "*layers.57.share_expert*",
920
+ "*layers.57.moe.share_expert*",
921
+ "*layers.57.moe.gate*",
922
+ "*layers.58.self_attn*",
923
+ "*layers.58.share_expert*",
924
+ "*layers.58.moe.share_expert*",
925
+ "*layers.58.moe.gate*",
926
+ "*layers.59.self_attn*",
927
+ "*layers.59.share_expert*",
928
+ "*layers.59.moe.share_expert*",
929
+ "*layers.59.moe.gate*",
930
+ "*layers.60.self_attn*",
931
+ "*layers.60.share_expert*",
932
+ "*layers.60.moe.share_expert*",
933
+ "*layers.60.moe.gate*",
934
+ "*layers.61.self_attn*",
935
+ "*layers.61.share_expert*",
936
+ "*layers.61.moe.share_expert*",
937
+ "*layers.61.moe.gate*",
938
+ "*layers.62.self_attn*",
939
+ "*layers.62.share_expert*",
940
+ "*layers.62.moe.share_expert*",
941
+ "*layers.62.moe.gate*",
942
+ "*layers.63.self_attn*",
943
+ "*layers.63.share_expert*",
944
+ "*layers.63.moe.share_expert*",
945
+ "*layers.63.moe.gate*",
946
+ "*layers.64.self_attn*",
947
+ "*layers.64.share_expert*",
948
+ "*layers.64.moe.share_expert*",
949
+ "*layers.64.moe.gate*",
950
+ "*layers.65.self_attn*",
951
+ "*layers.65.share_expert*",
952
+ "*layers.65.moe.share_expert*",
953
+ "*layers.65.moe.gate*",
954
+ "*layers.66.self_attn*",
955
+ "*layers.66.share_expert*",
956
+ "*layers.66.moe.share_expert*",
957
+ "*layers.66.moe.gate*",
958
+ "*layers.67.self_attn*",
959
+ "*layers.67.share_expert*",
960
+ "*layers.67.moe.share_expert*",
961
+ "*layers.67.moe.gate*",
962
+ "*layers.68.self_attn*",
963
+ "*layers.68.share_expert*",
964
+ "*layers.68.moe.share_expert*",
965
+ "*layers.68.moe.gate*",
966
+ "*layers.69.self_attn*",
967
+ "*layers.69.share_expert*",
968
+ "*layers.69.moe.share_expert*",
969
+ "*layers.69.moe.gate*",
970
+ "*layers.70.self_attn*",
971
+ "*layers.70.share_expert*",
972
+ "*layers.70.moe.share_expert*",
973
+ "*layers.70.moe.gate*",
974
+ "*layers.71.self_attn*",
975
+ "*layers.71.share_expert*",
976
+ "*layers.71.moe.share_expert*",
977
+ "*layers.71.moe.gate*",
978
+ "*layers.72.self_attn*",
979
+ "*layers.72.share_expert*",
980
+ "*layers.72.moe.share_expert*",
981
+ "*layers.72.moe.gate*",
982
+ "*layers.73.self_attn*",
983
+ "*layers.73.share_expert*",
984
+ "*layers.73.moe.share_expert*",
985
+ "*layers.73.moe.gate*",
986
+ "*layers.74.self_attn*",
987
+ "*layers.74.share_expert*",
988
+ "*layers.74.moe.share_expert*",
989
+ "*layers.74.moe.gate*",
990
+ "*layers.75.self_attn*",
991
+ "*layers.75.share_expert*",
992
+ "*layers.75.moe.share_expert*",
993
+ "*layers.75.moe.gate*",
994
+ "*layers.76.self_attn*",
995
+ "*layers.76.share_expert*",
996
+ "*layers.76.moe.share_expert*",
997
+ "*layers.76.moe.gate*",
998
+ "*layers.77.self_attn*",
999
+ "*layers.77.share_expert*",
1000
+ "*layers.77.moe.share_expert*",
1001
+ "*layers.77.moe.gate*",
1002
+ "*layers.78.self_attn*",
1003
+ "*layers.78.share_expert*",
1004
+ "*layers.78.moe.share_expert*",
1005
+ "*layers.78.moe.gate*",
1006
+ "*layers.79.self_attn*",
1007
+ "*layers.79.share_expert*",
1008
+ "*layers.79.moe.share_expert*",
1009
+ "*layers.79.moe.gate*",
1010
+ "*layers.80.self_attn*",
1011
+ "*layers.80.share_expert*",
1012
+ "*layers.80.moe.share_expert*",
1013
+ "*layers.80.moe.gate*",
1014
+ "*layers.81.self_attn*",
1015
+ "*layers.81.share_expert*",
1016
+ "*layers.81.moe.share_expert*",
1017
+ "*layers.81.moe.gate*",
1018
+ "*layers.82.self_attn*",
1019
+ "*layers.82.share_expert*",
1020
+ "*layers.82.moe.share_expert*",
1021
+ "*layers.82.moe.gate*",
1022
+ "*layers.83.self_attn*",
1023
+ "*layers.83.share_expert*",
1024
+ "*layers.83.moe.share_expert*",
1025
+ "*layers.83.moe.gate*",
1026
+ "*layers.84.self_attn*",
1027
+ "*layers.84.share_expert*",
1028
+ "*layers.84.moe.share_expert*",
1029
+ "*layers.84.moe.gate*",
1030
+ "*layers.85.self_attn*",
1031
+ "*layers.85.share_expert*",
1032
+ "*layers.85.moe.share_expert*",
1033
+ "*layers.85.moe.gate*",
1034
+ "*layers.86.self_attn*",
1035
+ "*layers.86.share_expert*",
1036
+ "*layers.86.moe.share_expert*",
1037
+ "*layers.86.moe.gate*",
1038
+ "*layers.87.self_attn*",
1039
+ "*layers.87.share_expert*",
1040
+ "*layers.87.moe.share_expert*",
1041
+ "*layers.87.moe.gate*",
1042
+ "*layers.88.self_attn*",
1043
+ "*layers.88.share_expert*",
1044
+ "*layers.88.moe.share_expert*",
1045
+ "*layers.88.moe.gate*",
1046
+ "*layers.88.moe.experts*",
1047
+ "*layers.89.self_attn*",
1048
+ "*layers.89.share_expert*",
1049
+ "*layers.89.moe.share_expert*",
1050
+ "*layers.89.moe.gate*",
1051
+ "*layers.89.moe.experts*",
1052
+ "*layers.90.self_attn*",
1053
+ "*layers.90.share_expert*",
1054
+ "*layers.90.moe.share_expert*",
1055
+ "*layers.90.moe.gate*",
1056
+ "*layers.90.moe.experts*",
1057
+ "*layers.91.self_attn*",
1058
+ "*layers.91.mlp*",
1059
+ "*layers.92.*",
1060
+ "*layers.93.*",
1061
+ "*layers.94.*",
1062
+ "vision_model*",
1063
+ "vit_large_projector*"
1064
+ ],
1065
+ "producer": {
1066
+ "name": "modelopt",
1067
+ "version": "0.47.0"
1068
+ }
1069
+ }
1070
+ }
hf_quant_config.json ADDED
@@ -0,0 +1,384 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "producer": {
3
+ "name": "modelopt",
4
+ "version": "0.47.0"
5
+ },
6
+ "quantization": {
7
+ "quant_algo": "NVFP4",
8
+ "kv_cache_quant_algo": null,
9
+ "group_size": 16,
10
+ "exclude_modules": [
11
+ "lm_head",
12
+ "model.embed_tokens",
13
+ "model.norm",
14
+ "*layers.0.self_attn*",
15
+ "*layers.0.mlp*",
16
+ "*layers.1.self_attn*",
17
+ "*layers.1.mlp*",
18
+ "*layers.2.self_attn*",
19
+ "*layers.2.mlp*",
20
+ "*layers.3.self_attn*",
21
+ "*layers.3.share_expert*",
22
+ "*layers.3.moe.share_expert*",
23
+ "*layers.3.moe.gate*",
24
+ "*layers.4.self_attn*",
25
+ "*layers.4.share_expert*",
26
+ "*layers.4.moe.share_expert*",
27
+ "*layers.4.moe.gate*",
28
+ "*layers.5.self_attn*",
29
+ "*layers.5.share_expert*",
30
+ "*layers.5.moe.share_expert*",
31
+ "*layers.5.moe.gate*",
32
+ "*layers.6.self_attn*",
33
+ "*layers.6.share_expert*",
34
+ "*layers.6.moe.share_expert*",
35
+ "*layers.6.moe.gate*",
36
+ "*layers.7.self_attn*",
37
+ "*layers.7.share_expert*",
38
+ "*layers.7.moe.share_expert*",
39
+ "*layers.7.moe.gate*",
40
+ "*layers.8.self_attn*",
41
+ "*layers.8.share_expert*",
42
+ "*layers.8.moe.share_expert*",
43
+ "*layers.8.moe.gate*",
44
+ "*layers.9.self_attn*",
45
+ "*layers.9.share_expert*",
46
+ "*layers.9.moe.share_expert*",
47
+ "*layers.9.moe.gate*",
48
+ "*layers.10.self_attn*",
49
+ "*layers.10.share_expert*",
50
+ "*layers.10.moe.share_expert*",
51
+ "*layers.10.moe.gate*",
52
+ "*layers.11.self_attn*",
53
+ "*layers.11.share_expert*",
54
+ "*layers.11.moe.share_expert*",
55
+ "*layers.11.moe.gate*",
56
+ "*layers.12.self_attn*",
57
+ "*layers.12.share_expert*",
58
+ "*layers.12.moe.share_expert*",
59
+ "*layers.12.moe.gate*",
60
+ "*layers.13.self_attn*",
61
+ "*layers.13.share_expert*",
62
+ "*layers.13.moe.share_expert*",
63
+ "*layers.13.moe.gate*",
64
+ "*layers.14.self_attn*",
65
+ "*layers.14.share_expert*",
66
+ "*layers.14.moe.share_expert*",
67
+ "*layers.14.moe.gate*",
68
+ "*layers.15.self_attn*",
69
+ "*layers.15.share_expert*",
70
+ "*layers.15.moe.share_expert*",
71
+ "*layers.15.moe.gate*",
72
+ "*layers.16.self_attn*",
73
+ "*layers.16.share_expert*",
74
+ "*layers.16.moe.share_expert*",
75
+ "*layers.16.moe.gate*",
76
+ "*layers.17.self_attn*",
77
+ "*layers.17.share_expert*",
78
+ "*layers.17.moe.share_expert*",
79
+ "*layers.17.moe.gate*",
80
+ "*layers.18.self_attn*",
81
+ "*layers.18.share_expert*",
82
+ "*layers.18.moe.share_expert*",
83
+ "*layers.18.moe.gate*",
84
+ "*layers.19.self_attn*",
85
+ "*layers.19.share_expert*",
86
+ "*layers.19.moe.share_expert*",
87
+ "*layers.19.moe.gate*",
88
+ "*layers.20.self_attn*",
89
+ "*layers.20.share_expert*",
90
+ "*layers.20.moe.share_expert*",
91
+ "*layers.20.moe.gate*",
92
+ "*layers.21.self_attn*",
93
+ "*layers.21.share_expert*",
94
+ "*layers.21.moe.share_expert*",
95
+ "*layers.21.moe.gate*",
96
+ "*layers.22.self_attn*",
97
+ "*layers.22.share_expert*",
98
+ "*layers.22.moe.share_expert*",
99
+ "*layers.22.moe.gate*",
100
+ "*layers.23.self_attn*",
101
+ "*layers.23.share_expert*",
102
+ "*layers.23.moe.share_expert*",
103
+ "*layers.23.moe.gate*",
104
+ "*layers.24.self_attn*",
105
+ "*layers.24.share_expert*",
106
+ "*layers.24.moe.share_expert*",
107
+ "*layers.24.moe.gate*",
108
+ "*layers.25.self_attn*",
109
+ "*layers.25.share_expert*",
110
+ "*layers.25.moe.share_expert*",
111
+ "*layers.25.moe.gate*",
112
+ "*layers.26.self_attn*",
113
+ "*layers.26.share_expert*",
114
+ "*layers.26.moe.share_expert*",
115
+ "*layers.26.moe.gate*",
116
+ "*layers.27.self_attn*",
117
+ "*layers.27.share_expert*",
118
+ "*layers.27.moe.share_expert*",
119
+ "*layers.27.moe.gate*",
120
+ "*layers.28.self_attn*",
121
+ "*layers.28.share_expert*",
122
+ "*layers.28.moe.share_expert*",
123
+ "*layers.28.moe.gate*",
124
+ "*layers.29.self_attn*",
125
+ "*layers.29.share_expert*",
126
+ "*layers.29.moe.share_expert*",
127
+ "*layers.29.moe.gate*",
128
+ "*layers.30.self_attn*",
129
+ "*layers.30.share_expert*",
130
+ "*layers.30.moe.share_expert*",
131
+ "*layers.30.moe.gate*",
132
+ "*layers.31.self_attn*",
133
+ "*layers.31.share_expert*",
134
+ "*layers.31.moe.share_expert*",
135
+ "*layers.31.moe.gate*",
136
+ "*layers.32.self_attn*",
137
+ "*layers.32.share_expert*",
138
+ "*layers.32.moe.share_expert*",
139
+ "*layers.32.moe.gate*",
140
+ "*layers.33.self_attn*",
141
+ "*layers.33.share_expert*",
142
+ "*layers.33.moe.share_expert*",
143
+ "*layers.33.moe.gate*",
144
+ "*layers.34.self_attn*",
145
+ "*layers.34.share_expert*",
146
+ "*layers.34.moe.share_expert*",
147
+ "*layers.34.moe.gate*",
148
+ "*layers.35.self_attn*",
149
+ "*layers.35.share_expert*",
150
+ "*layers.35.moe.share_expert*",
151
+ "*layers.35.moe.gate*",
152
+ "*layers.36.self_attn*",
153
+ "*layers.36.share_expert*",
154
+ "*layers.36.moe.share_expert*",
155
+ "*layers.36.moe.gate*",
156
+ "*layers.37.self_attn*",
157
+ "*layers.37.share_expert*",
158
+ "*layers.37.moe.share_expert*",
159
+ "*layers.37.moe.gate*",
160
+ "*layers.38.self_attn*",
161
+ "*layers.38.share_expert*",
162
+ "*layers.38.moe.share_expert*",
163
+ "*layers.38.moe.gate*",
164
+ "*layers.39.self_attn*",
165
+ "*layers.39.share_expert*",
166
+ "*layers.39.moe.share_expert*",
167
+ "*layers.39.moe.gate*",
168
+ "*layers.40.self_attn*",
169
+ "*layers.40.share_expert*",
170
+ "*layers.40.moe.share_expert*",
171
+ "*layers.40.moe.gate*",
172
+ "*layers.41.self_attn*",
173
+ "*layers.41.share_expert*",
174
+ "*layers.41.moe.share_expert*",
175
+ "*layers.41.moe.gate*",
176
+ "*layers.42.self_attn*",
177
+ "*layers.42.share_expert*",
178
+ "*layers.42.moe.share_expert*",
179
+ "*layers.42.moe.gate*",
180
+ "*layers.43.self_attn*",
181
+ "*layers.43.share_expert*",
182
+ "*layers.43.moe.share_expert*",
183
+ "*layers.43.moe.gate*",
184
+ "*layers.44.self_attn*",
185
+ "*layers.44.share_expert*",
186
+ "*layers.44.moe.share_expert*",
187
+ "*layers.44.moe.gate*",
188
+ "*layers.45.self_attn*",
189
+ "*layers.45.share_expert*",
190
+ "*layers.45.moe.share_expert*",
191
+ "*layers.45.moe.gate*",
192
+ "*layers.46.self_attn*",
193
+ "*layers.46.share_expert*",
194
+ "*layers.46.moe.share_expert*",
195
+ "*layers.46.moe.gate*",
196
+ "*layers.47.self_attn*",
197
+ "*layers.47.share_expert*",
198
+ "*layers.47.moe.share_expert*",
199
+ "*layers.47.moe.gate*",
200
+ "*layers.48.self_attn*",
201
+ "*layers.48.share_expert*",
202
+ "*layers.48.moe.share_expert*",
203
+ "*layers.48.moe.gate*",
204
+ "*layers.49.self_attn*",
205
+ "*layers.49.share_expert*",
206
+ "*layers.49.moe.share_expert*",
207
+ "*layers.49.moe.gate*",
208
+ "*layers.50.self_attn*",
209
+ "*layers.50.share_expert*",
210
+ "*layers.50.moe.share_expert*",
211
+ "*layers.50.moe.gate*",
212
+ "*layers.51.self_attn*",
213
+ "*layers.51.share_expert*",
214
+ "*layers.51.moe.share_expert*",
215
+ "*layers.51.moe.gate*",
216
+ "*layers.52.self_attn*",
217
+ "*layers.52.share_expert*",
218
+ "*layers.52.moe.share_expert*",
219
+ "*layers.52.moe.gate*",
220
+ "*layers.53.self_attn*",
221
+ "*layers.53.share_expert*",
222
+ "*layers.53.moe.share_expert*",
223
+ "*layers.53.moe.gate*",
224
+ "*layers.54.self_attn*",
225
+ "*layers.54.share_expert*",
226
+ "*layers.54.moe.share_expert*",
227
+ "*layers.54.moe.gate*",
228
+ "*layers.55.self_attn*",
229
+ "*layers.55.share_expert*",
230
+ "*layers.55.moe.share_expert*",
231
+ "*layers.55.moe.gate*",
232
+ "*layers.56.self_attn*",
233
+ "*layers.56.share_expert*",
234
+ "*layers.56.moe.share_expert*",
235
+ "*layers.56.moe.gate*",
236
+ "*layers.57.self_attn*",
237
+ "*layers.57.share_expert*",
238
+ "*layers.57.moe.share_expert*",
239
+ "*layers.57.moe.gate*",
240
+ "*layers.58.self_attn*",
241
+ "*layers.58.share_expert*",
242
+ "*layers.58.moe.share_expert*",
243
+ "*layers.58.moe.gate*",
244
+ "*layers.59.self_attn*",
245
+ "*layers.59.share_expert*",
246
+ "*layers.59.moe.share_expert*",
247
+ "*layers.59.moe.gate*",
248
+ "*layers.60.self_attn*",
249
+ "*layers.60.share_expert*",
250
+ "*layers.60.moe.share_expert*",
251
+ "*layers.60.moe.gate*",
252
+ "*layers.61.self_attn*",
253
+ "*layers.61.share_expert*",
254
+ "*layers.61.moe.share_expert*",
255
+ "*layers.61.moe.gate*",
256
+ "*layers.62.self_attn*",
257
+ "*layers.62.share_expert*",
258
+ "*layers.62.moe.share_expert*",
259
+ "*layers.62.moe.gate*",
260
+ "*layers.63.self_attn*",
261
+ "*layers.63.share_expert*",
262
+ "*layers.63.moe.share_expert*",
263
+ "*layers.63.moe.gate*",
264
+ "*layers.64.self_attn*",
265
+ "*layers.64.share_expert*",
266
+ "*layers.64.moe.share_expert*",
267
+ "*layers.64.moe.gate*",
268
+ "*layers.65.self_attn*",
269
+ "*layers.65.share_expert*",
270
+ "*layers.65.moe.share_expert*",
271
+ "*layers.65.moe.gate*",
272
+ "*layers.66.self_attn*",
273
+ "*layers.66.share_expert*",
274
+ "*layers.66.moe.share_expert*",
275
+ "*layers.66.moe.gate*",
276
+ "*layers.67.self_attn*",
277
+ "*layers.67.share_expert*",
278
+ "*layers.67.moe.share_expert*",
279
+ "*layers.67.moe.gate*",
280
+ "*layers.68.self_attn*",
281
+ "*layers.68.share_expert*",
282
+ "*layers.68.moe.share_expert*",
283
+ "*layers.68.moe.gate*",
284
+ "*layers.69.self_attn*",
285
+ "*layers.69.share_expert*",
286
+ "*layers.69.moe.share_expert*",
287
+ "*layers.69.moe.gate*",
288
+ "*layers.70.self_attn*",
289
+ "*layers.70.share_expert*",
290
+ "*layers.70.moe.share_expert*",
291
+ "*layers.70.moe.gate*",
292
+ "*layers.71.self_attn*",
293
+ "*layers.71.share_expert*",
294
+ "*layers.71.moe.share_expert*",
295
+ "*layers.71.moe.gate*",
296
+ "*layers.72.self_attn*",
297
+ "*layers.72.share_expert*",
298
+ "*layers.72.moe.share_expert*",
299
+ "*layers.72.moe.gate*",
300
+ "*layers.73.self_attn*",
301
+ "*layers.73.share_expert*",
302
+ "*layers.73.moe.share_expert*",
303
+ "*layers.73.moe.gate*",
304
+ "*layers.74.self_attn*",
305
+ "*layers.74.share_expert*",
306
+ "*layers.74.moe.share_expert*",
307
+ "*layers.74.moe.gate*",
308
+ "*layers.75.self_attn*",
309
+ "*layers.75.share_expert*",
310
+ "*layers.75.moe.share_expert*",
311
+ "*layers.75.moe.gate*",
312
+ "*layers.76.self_attn*",
313
+ "*layers.76.share_expert*",
314
+ "*layers.76.moe.share_expert*",
315
+ "*layers.76.moe.gate*",
316
+ "*layers.77.self_attn*",
317
+ "*layers.77.share_expert*",
318
+ "*layers.77.moe.share_expert*",
319
+ "*layers.77.moe.gate*",
320
+ "*layers.78.self_attn*",
321
+ "*layers.78.share_expert*",
322
+ "*layers.78.moe.share_expert*",
323
+ "*layers.78.moe.gate*",
324
+ "*layers.79.self_attn*",
325
+ "*layers.79.share_expert*",
326
+ "*layers.79.moe.share_expert*",
327
+ "*layers.79.moe.gate*",
328
+ "*layers.80.self_attn*",
329
+ "*layers.80.share_expert*",
330
+ "*layers.80.moe.share_expert*",
331
+ "*layers.80.moe.gate*",
332
+ "*layers.81.self_attn*",
333
+ "*layers.81.share_expert*",
334
+ "*layers.81.moe.share_expert*",
335
+ "*layers.81.moe.gate*",
336
+ "*layers.82.self_attn*",
337
+ "*layers.82.share_expert*",
338
+ "*layers.82.moe.share_expert*",
339
+ "*layers.82.moe.gate*",
340
+ "*layers.83.self_attn*",
341
+ "*layers.83.share_expert*",
342
+ "*layers.83.moe.share_expert*",
343
+ "*layers.83.moe.gate*",
344
+ "*layers.84.self_attn*",
345
+ "*layers.84.share_expert*",
346
+ "*layers.84.moe.share_expert*",
347
+ "*layers.84.moe.gate*",
348
+ "*layers.85.self_attn*",
349
+ "*layers.85.share_expert*",
350
+ "*layers.85.moe.share_expert*",
351
+ "*layers.85.moe.gate*",
352
+ "*layers.86.self_attn*",
353
+ "*layers.86.share_expert*",
354
+ "*layers.86.moe.share_expert*",
355
+ "*layers.86.moe.gate*",
356
+ "*layers.87.self_attn*",
357
+ "*layers.87.share_expert*",
358
+ "*layers.87.moe.share_expert*",
359
+ "*layers.87.moe.gate*",
360
+ "*layers.88.self_attn*",
361
+ "*layers.88.share_expert*",
362
+ "*layers.88.moe.share_expert*",
363
+ "*layers.88.moe.gate*",
364
+ "*layers.88.moe.experts*",
365
+ "*layers.89.self_attn*",
366
+ "*layers.89.share_expert*",
367
+ "*layers.89.moe.share_expert*",
368
+ "*layers.89.moe.gate*",
369
+ "*layers.89.moe.experts*",
370
+ "*layers.90.self_attn*",
371
+ "*layers.90.share_expert*",
372
+ "*layers.90.moe.share_expert*",
373
+ "*layers.90.moe.gate*",
374
+ "*layers.90.moe.experts*",
375
+ "*layers.91.self_attn*",
376
+ "*layers.91.mlp*",
377
+ "*layers.92.*",
378
+ "*layers.93.*",
379
+ "*layers.94.*",
380
+ "vision_model*",
381
+ "vit_large_projector*"
382
+ ]
383
+ }
384
+ }
model-extra-000.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5798d715bcae0a5cdd85e963ae0eb4f1c5bbbe4704fc24423c4b941ca12f8e1b
3
+ size 5466397616
model-extra-001.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d8965d16468a015de1d587a660e6b44d202b1253f1141535184e27e2f554c37
3
+ size 1677299136
model-layer-000.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dce1dfebe142c11ecc00aee3632a7573f2188b13b5b9ff87057f8042055b3a69
3
+ size 554208072
model-layer-001.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1e640e5a6d780325faf9469a21ff7e0cc0bc3370f237e56b4cb56c305818b7c1
3
+ size 554208072
model-layer-002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e0d41a65430b605422d082f9d1e87fe28bf206cd16019de17b9c253c7af47abb
3
+ size 554208072
model-layer-003.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2f353dfce8ee09c3ee7b8f91e3f506404b086219227949b4f53c2196572192b9
3
+ size 4030253168
model-layer-004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d42d0a4a6dea4c3c38e9474d520a8917f865a3c0eae1008c01525850a89a217a
3
+ size 3992238032
model-layer-005.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:60d8123ce6a64a4bba3ca1685aee1e4ae3e23ef0553100737373dc5ab306b2a9
3
+ size 3992238032
model-layer-006.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:643cf15d072d1bac1ff9e4710fc07b916a96a216cd83f43dbd70795e0a350f82
3
+ size 3992238032
model-layer-007.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6b2d23afc429640dcbb2bc3e3ef59aab024b428f4462c4b9743e121d4a7e6332
3
+ size 4030253168
model-layer-008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:99e77cfc79e3ea2773a2c9bd756a15ff024fe6c23ff4a36820aa8ae24a4e599c
3
+ size 3992238032
model-layer-009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7a4d84b5062ca545e8128bb0823b2608320d54e70c6c1bc4c736b2e6db2a1f3e
3
+ size 3992238032
model-layer-010.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3095ede9e5aeb62e68e518909d3b751b7dd493071535b4653d27b7c96b1f2658
3
+ size 3992238056
model-layer-011.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:102b9b9a149d9fa55de5a5949a0c3593062e629d0014bca0f04c8417eb10a4e5
3
+ size 4030253200
model-layer-012.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:406bf111c50354d2b15442b35cdbce40d21a5cb0ccbc9a9a8a594a7a9c0585a4
3
+ size 3992238056
model-layer-013.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e1c89907c1f6954abbdd73044d98a790235816f5aafe941792156ee330ac8239
3
+ size 3992238056
model-layer-014.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:38721f4f83a59c18b341e8c4ff6d90ae9919362775684411fda2b97161f221f4
3
+ size 3992238056
model-layer-015.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:11d012e3fd33540001a953297e65893e51871ec57c4841eaf718ace795080d1b
3
+ size 4030253200
model-layer-016.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d78acb313810ec483a52e717518940a61e81bd1422c9069c26f0fe07c47b1f2e
3
+ size 3992238056
model-layer-017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:303d7df94312a04254ab5c1c99567f08ac73036bc2c8555b9931146ce4e3bd95
3
+ size 3992238056
model-layer-018.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5bcef3cf91af47d01cccfbf66a4ebda584fda1c7a1220760a2cb9997181aaf94
3
+ size 3992238056
model-layer-019.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:48846cd2449c553fd36d6356f88a98106f44455595db5908cf3fa3a972cf3ce5
3
+ size 4030253200
model-layer-020.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2af7fc35dbd4ca516cbf5a4bbdd22776ad17fcfffe3dae33667e136a7c279d5a
3
+ size 3992238056
model-layer-021.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:69edfddb54ae974e8b441afab64dc206e0f2815b99b946bc33463a84abf1fee4
3
+ size 3992238056
model-layer-022.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8f699ac13783e616ccd5796ce730ef5e7f4523ad6c1e26b4843135ef547efc0f
3
+ size 3992238056
model-layer-023.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dfd3dd480637aec3ace46b38d95f1b88b95d056e6d30a3dde47f9a328e0a1d33
3
+ size 4030253200
model-layer-024.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:86067c89c66fe0c30582845d894b74fa4f7c193d039b931e79b3a120ea657294
3
+ size 3992238056
model-layer-025.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c9140bf3220418051f0d2902837972b897daa60c091b4ee9fa805cd256b580e0
3
+ size 3992238056
model-layer-026.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:08870fb96c5e55cdf4dac64d1caa80f2e6a4938302b9dcb24670f5436f88f1cf
3
+ size 3992238056
model-layer-027.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7a22518fb3c7e931db94e65d5448f170d57f80cc3d997a3fd5a81534fc1dfdce
3
+ size 4030253200
model-layer-028.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9a4ebe2e48ae6ecc3ee3cfbfdacfd1c8710684f51d79c0dd71c0604649be2852
3
+ size 3992238056
model-layer-029.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f63559c9dc6c7c855c10b7b7fab564d1725ba173b38b0cf8f6561a80247f6528
3
+ size 3992238056
model-layer-030.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c88b930d6a6414376e67908e8f8e9d2c37dc9338311e075122205114667b1bef
3
+ size 3992238056
model-layer-031.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:665f470193fceb71c67b536c9191697b9a35bb2c775104ac8e22cee187798b85
3
+ size 4030253200
model-layer-032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c69adc8e9e63dd5db2d6e0dba8aa268d531c5479945c37957ca82a8211edb344
3
+ size 3992238056
model-layer-033.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b4f411dd89049a82efdc5ffe9ce9fbe8533a7659e586ec35dc2329a39d6fcf99
3
+ size 3992238056
model-layer-034.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a1b8d02f96d21ee997b41a6cafd99d5c30bce9f6fba74832dc91c17ef7ba61ac
3
+ size 3992238056
model-layer-035.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fc9ec1882f7a2a9ddefc28431d6fdd433d8eed08df7a1edd8f0ca36bd5e758ca
3
+ size 4030253200
model-layer-036.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3f32a245c0801f6b823b90af9024457c6d2f1eead22c31c9c6f082b618538e58
3
+ size 3992238056
model-layer-037.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:59c173a7321479b6b9f876b0798e8bb7b4305b942347b30d039b9f5111aa9687
3
+ size 3992238056
model-layer-038.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:13bf83586b57409d55a59627551970df202616537cc350d6ce31ff2c81050b0a
3
+ size 3992238056
model-layer-039.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c95cc0c2f90c2ca88b1ec1e135ddf803396c20698d283006bfd2154c4332a7f6
3
+ size 4030253200
model-layer-040.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6cda0b2c873f70465cc65ac6c21911eeb86a931e09bb6bb5c38ce057d99e6641
3
+ size 3992238056