Upload folder using huggingface_hub

README.md

@@ -19,7 +19,7 @@ base_model: upstage/Solar-Open-100B
 
 # **Solar Open**
 
-**Solar Open** is Upstage's flagship **102B-parameter** large language model, trained **entirely from scratch** and released under the **Solar-Apache License 2.0** (see [LICENSE](./LICENSE)). As a **Mixture-of-Experts (MoE)** architecture, it delivers enterprise-grade performance in reasoning, instruction-following, and agentic capabilities—all while prioritizing transparency and customization for the open-source community.
+**Solar Open** is Upstage's flagship **102B-parameter** large language model, trained **entirely from scratch** and released under the **Solar-Apache License 2.0** (see [LICENSE](#LICENSE) for details). As a **Mixture-of-Experts (MoE)** architecture, it delivers enterprise-grade performance in reasoning, instruction-following, and agentic capabilities—all while prioritizing transparency and customization for the open-source community.
 
 ## Highlights
 
@@ -41,6 +41,18 @@ base_model: upstage/Solar-Open-100B
 * **Hardware Requirements:**
     * **Minimum:** 4x NVIDIA A100 (80GB)
 
+## License
+This repository contains both model weights and code,
+which are licensed under different terms:
+
+1. MODEL WEIGHTS (*.safetensors)
+   Licensed under **Solar-Apache License 2.0**
+   See: https://huggingface.co/upstage/Solar-Open-100B/blob/main/LICENSE
+
+2. CODE (*.py, *.json, *.jinja files)
+   Licensed under **Apache License 2.0**
+   See: https://www.apache.org/licenses/LICENSE-2.0
+
 ## Performance
 
 TBA
@@ -169,7 +181,7 @@ If you use Solar Open in your research, please cite:
 
 ```bibtex
 @misc{solar-open-2025,
-    title={Solar Open 100B},
+    title={Solar Open: Scaling Upstage's LLM Capabilities with MoE},
     author={Upstage AI},
     year={2025},
     url={https://huggingface.co/Upstage/Solar-Open-100B}

configuration_solar_open.py

@@ -1,17 +1,24 @@
 # coding=utf-8
-# Copyright 2025 Upstage AI. All rights reserved.
+# Copyright 2025 Upstage AI.
+# Copyright 2025 The ZhipuAI Inc. team and HuggingFace Inc. team.
 #
-# Licensed under the Solar-Apache License, Version 2.0 (the "License");
+# Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
 #
-#     https://huggingface.co/upstage/Solar-Open-100B/blob/main/LICENSE
+#     http://www.apache.org/licenses/LICENSE-2.0
 #
 # Unless required by applicable law or agreed to in writing, software
 # distributed under the License is distributed on an "AS IS" BASIS,
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+#
+# This file has been modified by Upstage AI including
+# -  Hyperparameter Adjustments: Modified the model architecture by increasing vocab_size and num_hidden_layers, while decreasing num_attention_heads, intermediate_size, and moe_intermediate_size.
+# RoPE Configuration: Replaced the generic rope_parameters argument with explicit rope_theta and rope_scaling parameters to define Rotary Positional Embeddings settings.
+#
+# Based on code from: https://github.com/huggingface/transformers/blob/main/src/transformers/models/glm4_moe/configuration_glm4_moe.py
 
 from transformers.configuration_utils import PretrainedConfig
 from transformers.modeling_rope_utils import rope_config_validation

generation_config.json

@@ -1,14 +1,14 @@
 {
   "_from_model_config": true,
   "bos_token_id": 1,
+  "do_sample": true,
   "eos_token_id": [
     2,
     24,
     25
   ],
   "pad_token_id": 2,
-  "transformers_version": "4.57.3",
-  "do_sample": true,
   "temperature": 0.8,
-  "top_p": 0.95
-}
+  "top_p": 0.95,
+  "transformers_version": "4.57.3"
+}

model-00001-of-00013.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:150e9ab222ca9289a3bd5aecae07cd27b66ca7eec8b5a21d2c802c13602efc38
+oid sha256:9260bbc8e6e1425492b6dd96306f607add675907b05383e6de3b44acaa8d8929
 size 5000328728

model-00002-of-00013.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c1e1abe541bbd9b4c30bcf9614a9d7c05b6c78ca839f2a2b129a2aa70ab85c48
+oid sha256:835d751936c061113ab8583cef82751df39147fcf14e2b0faf67b2d945807044
 size 4998530632

model-00003-of-00013.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5a90f1d90c44d63fe368f471307c6670b01e01e4adb09d95d562933471b8c341
+oid sha256:8d8f5e4a8d07e2e988f351009b0adeee7bf8674267e30ca332551990d8050c61
 size 4999786408

model-00004-of-00013.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cad0c14c7daca53028b97e253e1695789587a9de0dbc6a3cc5a562c3716fa487
+oid sha256:84a4661748c1515a45699d38f28d0884d1c4b24af63e7aca18d41caa41c075cb
 size 4998535288

model-00005-of-00013.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dec14a7e564366db35f1c27dd2beaea6038498d13e204d899d3dc41901530f87
+oid sha256:4ab023431712224e1600f109be0eab7e8ab406728d16d13f5341b00a047065f0
 size 4998535320

model-00006-of-00013.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:74a8329165658e72761caeabd50ac33257e9187faf7b5db4812599af00e7d996
+oid sha256:9d5928333173e3b057656a2196c6dcf3ce49ca04d3506fb5ee2815b4bba611f8
 size 4998535320

model-00007-of-00013.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:83db4ea3a9d055c80e37e841955835f238f5ccddde0a8ee244febe63d4cd4781
+oid sha256:a5a683a150b4343a9693a9c5fba24d30395b6950da65e5d04cc9f3c3be75c9bc
 size 4998535320

model-00008-of-00013.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7a21b173a46c1181df522e558d097512343614c901ccb84d6de93167f93359b1
+oid sha256:6757548ed17bdc777b3aaa8d3f24d1c6591ef21fe1da109c025febbe890ddff7
 size 4998535336

model-00009-of-00013.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aef3cf4745e8607e1ad00ffa402de44384847482261ab4b5dc95acdb7b48fe6a
+oid sha256:5a1ce865921c31a474a44f4698f89a35976e60f28d8e3bf173bf0775e8fc9e7b
 size 4998535488

model-00010-of-00013.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e1d3ba0f7d427997a6bb4af69dc6a2ab2fc1990921ee2c5287e337528ac7543d
+oid sha256:9d41300d6e6fe3eb50aee95e48b19a43f3fdf6613e72fbf0e95233b064004242
 size 4999790016

model-00011-of-00013.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7649d0e3eb6474fc3845f3351ef6aea75ee993c6c14086a2614759666e8b1e3e
+oid sha256:2751a64b426a9795c7d04a19b9ece9bb2495afb96ca6d9093546a109a39e8599
 size 4998535264

model-00012-of-00013.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ae0186b1d258e8801538c8b3c0d9e4c3eeda1fd20d8da32bca6d765c9b8ba353
+oid sha256:1099e321bad6a5235e5de4a2961fecede247f4f0f7d1e5d88db9f3e3b45e1878
 size 4587386184

modeling_solar_open.py

@@ -1,17 +1,25 @@
 # coding=utf-8
-# Copyright 2025 Upstage AI. All rights reserved.
+# Copyright 2025 Upstage AI.
+# Copyright 2025 The GLM4 & ZhipuAI team and HuggingFace Inc. team.
 #
-# Licensed under the Solar-Apache License, Version 2.0 (the "License");
+# Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
 #
-#     https://huggingface.co/upstage/Solar-Open-100B/blob/main/LICENSE
+#     http://www.apache.org/licenses/LICENSE-2.0
 #
 # Unless required by applicable law or agreed to in writing, software
 # distributed under the License is distributed on an "AS IS" BASIS,
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+#
+# This file has been modified by Upstage AI including:
+# - Hybrid MoE Architecture: Replaced the standard dense structure with a depth-dependent Hybrid MoE, adding `SolarOpenMoE` and `SolarOpenTopkRouter` classes.
+# - RoPE Strategy: Changed the rotary position embedding strategy from GLM4's interleaved rotation to Llama-style block rotation (via modified `rotate_half`).
+# - Normalization Logic: Simplified the layer normalization structure by removing GLM4's extra post-operation norms and adding optional Query-Key Normalization (`use_qk_norm`).
+#
+# Based on code from: https://github.com/huggingface/transformers/blob/main/src/transformers/models/glm4/modeling_glm4.py
 
 from typing import Callable, Optional, Union
 

parallel_tool_call_logits_processor.py

@@ -1,11 +1,11 @@
 # coding=utf-8
-# Copyright 2025 Upstage AI. All rights reserved.
+# Copyright 2025 Upstage AI.
 #
-# Licensed under the Solar-Apache License, Version 2.0 (the "License");
+# Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
 #
-#     https://huggingface.co/upstage/Solar-Open-100B/blob/main/LICENSE
+#     http://www.apache.org/licenses/LICENSE-2.0
 #
 # Unless required by applicable law or agreed to in writing, software
 # distributed under the License is distributed on an "AS IS" BASIS,

solar_open_logits_processor.py

@@ -1,11 +1,11 @@
 # coding=utf-8
-# Copyright 2025 Upstage AI. All rights reserved.
+# Copyright 2025 Upstage AI.
 #
-# Licensed under the Solar-Apache License, Version 2.0 (the "License");
+# Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
 #
-#     https://huggingface.co/upstage/Solar-Open-100B/blob/main/LICENSE
+#     http://www.apache.org/licenses/LICENSE-2.0
 #
 # Unless required by applicable law or agreed to in writing, software
 # distributed under the License is distributed on an "AS IS" BASIS,

solar_open_reasoning_parser.py

@@ -1,11 +1,11 @@
 # coding=utf-8
-# Copyright 2025 Upstage AI. All rights reserved.
+# Copyright 2025 Upstage AI.
 #
-# Licensed under the Solar-Apache License, Version 2.0 (the "License");
+# Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
 #
-#     https://huggingface.co/upstage/Solar-Open-100B/blob/main/LICENSE
+#     http://www.apache.org/licenses/LICENSE-2.0
 #
 # Unless required by applicable law or agreed to in writing, software
 # distributed under the License is distributed on an "AS IS" BASIS,

solar_open_tool_parser.py

@@ -1,11 +1,11 @@
 # coding=utf-8
-# Copyright 2025 Upstage AI. All rights reserved.
+# Copyright 2025 Upstage AI.
 #
-# Licensed under the Solar-Apache License, Version 2.0 (the "License");
+# Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
 #
-#     https://huggingface.co/upstage/Solar-Open-100B/blob/main/LICENSE
+#     http://www.apache.org/licenses/LICENSE-2.0
 #
 # Unless required by applicable law or agreed to in writing, software
 # distributed under the License is distributed on an "AS IS" BASIS,

tokenizer_config.json

@@ -36751,12 +36751,14 @@
   "clean_up_tokenization_spaces": false,
   "eos_token": "<|endoftext|>",
   "extra_special_tokens": {},
-  "model_input_names": ["input_ids", "attention_mask"],
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ],
   "model_max_length": 1000000000000000019884624838656,
   "pad_token": "<|endoftext|>",
   "padding_side": "left",
   "split_special_tokens": false,
   "tokenizer_class": "PreTrainedTokenizerFast",
-  "unk_token": "<unk>",
-  "chat_template": "{#- ======== Template Parameters ========  #}\r\n{%- set add_generation_prompt = add_generation_prompt if add_generation_prompt is defined else true %}\r\n{%- set default_system_prompt = default_system_prompt if default_system_prompt is defined else true %}\r\n{%- set reasoning_effort = reasoning_effort if reasoning_effort is defined else \"high\" %}\r\n{%- set think_render_option = think_render_option if think_render_option is defined else \"lastthink\" %}\r\n\r\n{#- ======== System Block State ========  #}\r\n{%- set sys_ns = namespace(is_first_block=true) -%}\r\n\r\n{#- ======== Find last user message index ========  #}\r\n{%- set last_user_idx = namespace(value=-1) -%}\r\n{%- for message in messages -%}\r\n    {%- if message.role == 'user' -%}\r\n        {%- set last_user_idx.value = loop.index0 -%}\r\n    {%- endif -%}\r\n{%- endfor -%}\r\n\r\n{#- ======== System messages renderers ========  #}\r\n{%- macro render_system_message(user_system_messages) %}\r\n    {%- if default_system_prompt %}\r\n        {%- if not sys_ns.is_first_block %}{{- \"\\n\\n\" }}{%- endif %}\r\n        {%- set sys_ns.is_first_block = false %}\r\n        {{- \"## Provider System Prompt\\n\\nYou are Solar Open 100B, a large language model trained by Upstage AI, a Korean startup. Your knowledge cutoff is 2025-07. The current date is \" + strftime_now(\"%Y-%m-%d\") + \".\" }}\r\n    {%- endif -%}\r\n    {%- if user_system_messages %}\r\n        {%- if not sys_ns.is_first_block %}{{- \"\\n\\n\" }}{%- endif %}\r\n        {%- set sys_ns.is_first_block = false %}\r\n        {{- \"## System Prompt\" }}\r\n        {%- for system_message in user_system_messages %}\r\n            {{- \"\\n\\n\" }}\r\n            {{- system_message }}\r\n        {%- endfor %}\r\n    {%- endif -%}\r\n{%- endmacro %}\r\n\r\n{%- macro render_tool_instruction(tools) %}\r\n    {%- if not sys_ns.is_first_block %}{{- \"\\n\\n\" }}{%- endif %}\r\n    {%- set sys_ns.is_first_block = false %}\r\n    {{- \"## Tools\\n\\n### Tool Call Instruction\" }}\r\n    {{- \"\\nYou may invoke one or more tools to assist with the user's query. Available tools are provided in JSON Schema format: <|tools:begin|><|tool:begin|><tools-json-object><|tool:end|>...<|tools:end|>\\n\" }}\r\n    {{- \"\\n### Available Tools\\n\" }}\r\n    {{- \"<|tools:begin|>\" }}\r\n    {%- for tool in tools %}\r\n        {{- \"<|tool:begin|>\" }}\r\n        {{- tool.function | tojson }}\r\n        {{- \"<|tool:end|>\" }}\r\n    {%- endfor %}\r\n    {{- \"<|tools:end|>\\n\" }}\r\n    {{- \"\\n### Tool Call Format\\n\" }}\r\n    {{- \"For each tool call, return a JSON object with the following structure, enclosed within <|tool_call:begin|> and <|tool_call:end|> tags: \\n<|tool_call:begin|><tool-call-id><|tool_call:name|><tool-name><|tool_call:args|><args-json-object><|tool_call:end|>\\n\" }}\r\n    {{- \"- The <tool-call-id> must be a randomly generated string consisting of 10 lowercase letters (a-z) and\/or digits (0-9) (e.g., a1b2c3d4e5)\\n\" }}\r\n    {{- \"\\n### Tool Response Format\\n\" }}\r\n    {{- \"Each tool is responded by `tool` with the following structure:\\n<|tool_response:id|><tool-call-id><|tool_response:name|><tool-name><|tool_response:result|><results><|tool_response:end|>\\n\" }}\r\n    {{- \"- Ensure the <tool-call-id> matches the corresponding tool call\" -}}\r\n{%- endmacro %}\r\n\r\n{%- macro render_json_response_format_instruction(response_format) %}\r\n    {%- if not sys_ns.is_first_block %}{{- \"\\n\\n\" }}{%- endif %}\r\n    {%- set sys_ns.is_first_block = false %}\r\n    {{- \"## Output Format Constraint\" }}\r\n    {{- \"\\n\\nYour final response should follow the JSON schema: \\n[Start of schema]\" }}\r\n    {{- response_format }}\r\n    {{- \"\\n[End of schema]\\nPlease ensure your answers adhere to this format and do not contain any unnecessary text.\" }}\r\n{%- endmacro %}\r\n\r\n{%- macro get_tool_name(messages, tool_call_id) %}\r\n    {%- for msg in messages -%}\r\n        {%- if msg.role == 'assistant' and msg.tool_calls -%}\r\n            {%- for tool_call in msg.tool_calls -%}\r\n                {%- if tool_call.id == tool_call_id -%}\r\n                    {{- tool_call.function.name }}\r\n                {%- endif -%}\r\n            {%- endfor -%}\r\n        {%- endif -%}\r\n    {%- endfor -%}\r\n{%- endmacro %}\r\n\r\n{%- macro render_tool_arguments(tool_arguments) %}\r\n    {%- if tool_arguments is mapping -%}\r\n        {{- tool_arguments | tojson }}\r\n    {%- else -%}\r\n        {{- tool_arguments }}\r\n    {%- endif -%}\r\n{%- endmacro %}\r\n\r\n{#- ======== Render system message ========  #}\r\n{%- set ns = namespace(system_messages=[]) -%}\r\n{%- for message in messages -%}\r\n    {%- if message.role == 'system' -%}\r\n        {%- set ns.system_messages = ns.system_messages + [message.content] -%}\r\n    {%- endif -%}\r\n{%- endfor -%}\r\n\r\n{%- if ns.system_messages or default_system_prompt or tools or response_format -%}\r\n    {{- \"<|begin|>system<|content|>\" }}\r\n        {{- render_system_message(ns.system_messages) }}\r\n        {%- if tools -%}\r\n            {{- render_tool_instruction(tools) }}\r\n        {%- endif %}\r\n        {%- if response_format -%}\r\n            {{- render_json_response_format_instruction(response_format) }}\r\n        {%- endif %}\r\n    {{- \"<|end|>\" }}\r\n{%- endif -%}\r\n\r\n{#- ======== Render main messages ========  #}\r\n{%- for message in messages -%}\r\n    {%- if message.role == 'user' -%}\r\n         {{- \"<|begin|>user<|content|>\" + message.content + \"<|end|>\" }}\r\n    {%- elif message.role == 'tool' -%}\r\n        {%- set prev_is_tool = loop.index0 > 0 and messages[loop.index0 - 1].role == 'tool' -%}\r\n        {%- set next_is_tool = loop.index0 < (messages | length - 1) and messages[loop.index0 + 1].role == 'tool' -%}\r\n        {%- if not prev_is_tool -%}\r\n            {{- \"<|begin|>tool<|tool_response|>\" }}\r\n        {%- endif -%}\r\n        {{- \"<|tool_response:begin|>\" + message.tool_call_id + \"<|tool_response:name|>\" }}\r\n        {{- get_tool_name(messages, message.tool_call_id) }}\r\n        {{- \"<|tool_response:result|>\" }}\r\n        {{- message.content }}\r\n        {{- \"<|tool_response:end|>\" }}\r\n        {%- if not next_is_tool -%}\r\n            {{- \"<|end|>\" }}\r\n        {%- endif -%}\r\n    {%- elif message.role == 'assistant' -%}\r\n        {#- ======== Assistant Thinking ========  #}\r\n        {%- if think_render_option == \"all\" -%}\r\n            {%- if message.reasoning -%}\r\n                {{- \"<|begin|>assistant<|think|>\" + message.reasoning + \"<|end|>\" }}\r\n            {%- endif -%}\r\n        {%- elif think_render_option == \"lastthink\" -%}\r\n            {%- if message.reasoning and loop.index0 > last_user_idx.value -%}\r\n                {{- \"<|begin|>assistant<|think|>\" + message.reasoning + \"<|end|>\" }}\r\n            {%- endif -%}\r\n        {%- endif -%}\r\n\r\n        {#- ======== Assistant Messages ========  #}\r\n        {%- if message.tool_calls -%}\r\n            {{- \"<|begin|>assistant<|tool_calls|>\" }}\r\n            {%- for tool_call in message.tool_calls -%}\r\n                {{- \"<|tool_call:begin|>\" + tool_call.id +\"<|tool_call:name|>\" + tool_call.function.name + \"<|tool_call:args|>\" }}\r\n                {{- render_tool_arguments(tool_call.function.arguments) }}\r\n                {{- \"<|tool_call:end|>\" }}\r\n            {%- endfor -%}\r\n            {{- \"<|calls|>\" }}\r\n        {%- else -%}\r\n            {{- \"<|begin|>assistant<|content|>\" + message.content + \"<|end|>\" }}\r\n        {%- endif -%}\r\n    {%- endif -%}\r\n{%- endfor -%}\r\n\r\n{%- if add_generation_prompt -%}\r\n    {%- if reasoning_effort in [\"low\", \"minimal\"] -%}\r\n        {{- \"<|begin|>assistant<|think|><|end|>\" }}\r\n    {%- endif -%}\r\n    {{- \"<|begin|>assistant\" }}\r\n{%- endif -%}"
+  "unk_token": "<unk>"
 }