{"_id":"69b063ae64c5610da7d4114b","id":"nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","private":false,"pipeline_tag":"text-generation","library_name":"transformers","tags":["transformers","safetensors","nemotron_h","text-generation","nvidia","pytorch","nemotron-3","latent-moe","mtp","conversational","custom_code","en","fr","es","it","de","ja","zh","dataset:nvidia/nemotron-post-training-v3","dataset:nvidia/nemotron-pre-training-datasets","arxiv:2512.20848","arxiv:2512.20856","license:other","eval-results","endpoints_compatible","deploy:sagemaker","deploy:azure","region:us"],"downloads":1309317,"likes":425,"modelId":"nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","author":"nvidia","sha":"2dc98e2afe4face0e4ce40972a915c45368bd34a","lastModified":"2026-08-25T15:48:50.000Z","gated":false,"disabled":false,"widgetData":[{"text":"Hi, what can you help me with?"},{"text":"What is 84 * 3 / 2?"},{"text":"Tell me an interesting fact about the universe!"},{"text":"Explain quantum computing in simple terms."}],"model-index":null,"config":{"architectures":["NemotronHForCausalLM"],"auto_map":{"AutoConfig":"configuration_nemotron_h.NemotronHConfig","AutoModelForCausalLM":"modeling_nemotron_h.NemotronHForCausalLM"},"model_type":"nemotron_h","num_experts_per_tok":22,"tokenizer_config":{"bos_token":"<s>","eos_token":"<|im_end|>","pad_token":"<|im_end|>","unk_token":"<unk>"},"chat_template_jinja":"{% macro render_extra_keys(json_dict, handled_keys) %}\n    {%- if json_dict is mapping %}\n        {%- for json_key in json_dict if json_key not in handled_keys %}\n            {%- if json_dict[json_key] is mapping or (json_dict[json_key] is sequence and json_dict[json_key] is not string) %}\n                {{- '\\n<' ~ json_key ~ '>' ~ (json_dict[json_key] | tojson | safe) ~ '</' ~ json_key ~ '>' }}\n            {%- else %}\n                {{-'\\n<' ~ json_key ~ '>' ~ (json_dict[json_key] | string) ~ '</' ~ json_key ~ '>' }}\n            {%- endif %}\n        {%- endfor %}\n    {%- endif %}\n{% endmacro %}\n{%- set enable_thinking = enable_thinking if enable_thinking is defined else True %}\n{%- set low_effort = low_effort if low_effort is defined else False %}\n{%- set truncate_history_thinking = truncate_history_thinking if truncate_history_thinking is defined else True %}\n\n{%- set ns = namespace(last_user_idx = -1) %}\n{%- set loop_messages = messages %}\n{%- for m in loop_messages %}\n  {%- if m[\"role\"] == \"user\" %}\n    {%- set ns.last_user_idx = loop.index0 %}\n  {%- endif %}\n{%- endfor %}\n\n{%- if messages[0][\"role\"] == \"system\" %}\n    {%- set system_message = messages[0][\"content\"] %}\n    {%- set loop_messages = messages[1:] %}\n{%- else %}\n    {%- set system_message = \"\" %}\n    {%- set loop_messages = messages %}\n{%- endif %}\n{%- if not tools is defined %}\n    {%- set tools = [] %}\n{%- endif %}\n{# Recompute last_user_idx relative to loop_messages after handling system #}\n{%- set ns = namespace(last_user_idx = -1) %}\n{%- for m in loop_messages %}\n  {%- if m[\"role\"] == \"user\" %}\n    {%- set ns.last_user_idx = loop.index0 %}\n  {%- endif %}\n{%- endfor %}\n{%- if system_message is defined %}\n    {{- \"<|im_start|>system\\n\" + system_message }}\n{%- else %}\n    {%- if tools is iterable and tools | length > 0 %}\n        {{- \"<|im_start|>system\\n\" }}\n    {%- endif %}\n{%- endif %}\n{%- if tools is iterable and tools | length > 0 %}\n    {%- if system_message is defined and system_message | length > 0 %}\n        {{- \"\\n\\n\" }}\n    {%- endif %}\n    {{- \"# Tools\\n\\nYou have access to the following functions:\\n\\n\" }}\n    {{- \"<tools>\" }}\n    {%- for tool in tools %}\n        {%- if tool.function is defined %}\n            {%- set tool = tool.function %}\n        {%- endif %}\n        {{- \"\\n<function>\\n<name>\" ~ tool.name ~ \"</name>\" }}\n        {%- if tool.description is defined %}\n            {{- '\\n<description>' ~ (tool.description | trim) ~ '</description>' }}\n        {%- endif %}\n        {{- '\\n<parameters>' }}\n        {%- if tool.parameters is defined and tool.parameters is mapping and tool.parameters.properties is defined and tool.parameters.properties is mapping %}\n            {%- for param_name, param_fields in tool.parameters.properties|items %}\n                {{- '\\n<parameter>' }}\n                {{- '\\n<name>' ~ param_name ~ '</name>' }}\n                {%- if param_fields.type is defined %}\n                    {{- '\\n<type>' ~ (param_fields.type | string) ~ '</type>' }}\n                {%- endif %}\n                {%- if param_fields.description is defined %}\n                    {{- '\\n<description>' ~ (param_fields.description | trim) ~ '</description>' }}\n                {%- endif %}\n                {%- if param_fields.enum is defined %}\n                    {{- '\\n<enum>' ~ (param_fields.enum | tojson | safe) ~ '</enum>' }}\n                {%- endif %}\n                {%- set handled_keys = ['name', 'type', 'description', 'enum'] %}\n                {{- render_extra_keys(param_fields, handled_keys) }}\n                {{- '\\n</parameter>' }}\n            {%- endfor %}\n        {%- endif %}\n        {% set handled_keys = ['type', 'properties', 'required'] %}\n        {{- render_extra_keys(tool.parameters, handled_keys) }}\n        {%- if tool.parameters is defined and tool.parameters.required is defined %}\n            {{- '\\n<required>' ~ (tool.parameters.required | tojson | safe) ~ '</required>' }}\n        {%- endif %}\n        {{- '\\n</parameters>' }}\n        {%- set handled_keys = ['type', 'name', 'description', 'parameters'] %}\n        {{- render_extra_keys(tool, handled_keys) }}\n        {{- '\\n</function>' }}\n    {%- endfor %}\n    {{- \"\\n</tools>\" }}\n\n    {{- '\\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=...></function> block must be nested within <tool_call></tool_call> XML tags\\n- Required parameters MUST be specified\\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\\n</IMPORTANT>' }}\n{%- endif %}\n\n\n{%- if system_message is defined %}\n    {{- '<|im_end|>\\n' }}\n{%- else %}\n    {%- if tools is iterable and tools | length > 0 %}\n        {{- '<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n\n{%- for message in loop_messages %}\n    {%- if message.role == \"assistant\" %}\n        {# Add reasoning content in to content field for unified processing below. #}\n        {%- if message.reasoning_content is defined and message.reasoning_content is string and message.reasoning_content | trim | length > 0 %}\n            {%- set content = \"<think>\\n\" ~ message.reasoning_content ~ \"\\n</think>\\n\" ~ (message.content | default('', true)) %}\n        {%- else %}\n            {%- set content = message.content | default('', true) %}\n            {%- if content is string -%}\n                {# Allow downstream logic to to take care of broken thought, only handle coherent reasoning here. #}\n                {%- if '<think>' not in content and '</think>' not in content -%}\n                    {%- set content = \"<think></think>\" ~ content -%}\n                {%- endif -%}\n            {%- else -%}\n                {%- set content = content -%}\n            {%- endif -%}\n        {%- endif %}\n        {%- if message.tool_calls is defined and message.tool_calls is iterable and message.tool_calls | length > 0 %}\n            {# Assistant message has tool calls. #}\n            {{- '<|im_start|>assistant\\n' }}\n                {%- set include_content = not (truncate_history_thinking and loop.index0 < ns.last_user_idx) %}\n                {%- if content is string and content | trim | length > 0 %}\n                    {%- if include_content %}\n                        {{- (content | trim) ~ '\\n' -}}\n                    {%- else %}\n                        {%- set c = (content | string) %}\n                        {%- if '</think>' in c %}\n                            {# Keep only content after the last closing think. Also generation prompt causes this. #}\n                            {%- set c = c.split('</think>')[-1] %}\n                        {%- elif '<think>' in c %}\n                            {# If <think> was opened but never closed, drop the trailing think segment #}\n                            {%- set c = c.split('<think>')[0] %}\n                        {%- endif %}\n                        {%- set c = \"<think></think>\" ~ c | trim %}\n                        {%- if c | length > 0 %}\n                            {{- c ~ '\\n' -}}\n                        {%- endif %}\n                    {%- endif %}\n                {%- else %}\n                    {{- \"<think></think>\" -}}\n                {%- endif %}\n                {%- for tool_call in message.tool_calls %}\n                    {%- if tool_call.function is defined %}\n                        {%- set tool_call = tool_call.function %}\n                    {%- endif %}\n                    {{- '<tool_call>\\n<function=' ~ tool_call.name ~ '>\\n' -}}\n                        {%- if tool_call.arguments is defined %}\n                            {%- for args_name, args_value in tool_call.arguments|items %}\n                                {{- '<parameter=' ~ args_name ~ '>\\n' -}}\n                                    {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}\n                                {{- args_value ~ '\\n</parameter>\\n' -}}\n                            {%- endfor %}\n                        {%- endif %}\n                    {{- '</function>\\n</tool_call>\\n' -}}\n                {%- endfor %}\n                {{- '<|im_end|>\\n' }}\n        {%- else %}\n            {# Assistant message doesn't have tool calls. #}\n            {%- if not (truncate_history_thinking and loop.index0 < ns.last_user_idx) %}\n                {{- '<|im_start|>assistant\\n' ~ (content | default('', true) | string | trim) ~ '<|im_end|>\\n' }}\n            {%- else %}\n                {%- set c = (content | default('', true) | string) %}\n                {%- if '<think>' in c and '</think>' in c %}\n                    {%- set c = \"<think></think>\" ~ c.split('</think>')[-1] %}\n                {%- endif %}\n                {%- set c = c | trim %}\n                {%- if c | length > 0 %}\n                    {{- '<|im_start|>assistant\\n' ~ c ~ '<|im_end|>\\n' }}\n                {%- else %}\n                    {{- '<|im_start|>assistant\\n<|im_end|>\\n' }}\n                {%- endif %}\n            {%- endif %}\n        {%- endif %}\n    {%- elif message.role == \"user\" or message.role == \"system\" %}\n        {{- '<|im_start|>' + message.role + '\\n' }}\n        {%- set content = message.content | string %}\n        {%- if message.role == \"user\" and loop.index0 == ns.last_user_idx and low_effort %}\n            {{- content + '\\n\\n{reasoning effort: low}' }}\n        {%- else %}\n            {{- content }}\n        {%- endif %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if loop.previtem and loop.previtem.role != \"tool\" %}\n            {{- '<|im_start|>user\\n' }}\n        {%- endif %}\n        {{- '<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>\\n' }}\n        {%- if not loop.last and loop.nextitem.role != \"tool\" %}\n            {{- '<|im_end|>\\n' }}\n        {%- elif loop.last %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- else %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>\\n' }}\n    {%- endif %}\n{%- endfor %}\n\n{%- if add_generation_prompt %}\n    {%- if enable_thinking %}\n        {{- '<|im_start|>assistant\\n<think>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>assistant\\n<think></think>' }}\n    {%- endif %}\n{%- endif %}\n"},"cardData":{"library_name":"transformers","license":"other","license_name":"nvidia-nemotron-open-model-license","license_link":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/","pipeline_tag":"text-generation","language":["en","fr","es","it","de","ja","zh"],"tags":["nvidia","pytorch","nemotron-3","latent-moe","mtp"],"datasets":["nvidia/nemotron-post-training-v3","nvidia/nemotron-pre-training-datasets"],"track_downloads":true},"transformersInfo":{"auto_model":"AutoModelForCausalLM","pipeline_tag":"text-generation","processor":"AutoTokenizer"},"siblings":[{"rfilename":".eval_results/gpqa.yaml"},{"rfilename":".eval_results/gpqa_with_tools.yaml"},{"rfilename":".eval_results/hle.yaml"},{"rfilename":".eval_results/hle_with_tools.yaml"},{"rfilename":".eval_results/mmlu_pro.yaml"},{"rfilename":".eval_results/swe_bench_verified.yaml"},{"rfilename":".eval_results/terminal-bench_2.0.yaml"},{"rfilename":".gitattributes"},{"rfilename":"README.md"},{"rfilename":"__init__.py"},{"rfilename":"accuracy_chart.png"},{"rfilename":"bias.md"},{"rfilename":"chat_template.jinja"},{"rfilename":"config.json"},{"rfilename":"configuration_nemotron_h.py"},{"rfilename":"explainability.md"},{"rfilename":"generation_config.json"},{"rfilename":"model-00001-of-00050.safetensors"},{"rfilename":"model-00002-of-00050.safetensors"},{"rfilename":"model-00003-of-00050.safetensors"},{"rfilename":"model-00004-of-00050.safetensors"},{"rfilename":"model-00005-of-00050.safetensors"},{"rfilename":"model-00006-of-00050.safetensors"},{"rfilename":"model-00007-of-00050.safetensors"},{"rfilename":"model-00008-of-00050.safetensors"},{"rfilename":"model-00009-of-00050.safetensors"},{"rfilename":"model-00010-of-00050.safetensors"},{"rfilename":"model-00011-of-00050.safetensors"},{"rfilename":"model-00012-of-00050.safetensors"},{"rfilename":"model-00013-of-00050.safetensors"},{"rfilename":"model-00014-of-00050.safetensors"},{"rfilename":"model-00015-of-00050.safetensors"},{"rfilename":"model-00016-of-00050.safetensors"},{"rfilename":"model-00017-of-00050.safetensors"},{"rfilename":"model-00018-of-00050.safetensors"},{"rfilename":"model-00019-of-00050.safetensors"},{"rfilename":"model-00020-of-00050.safetensors"},{"rfilename":"model-00021-of-00050.safetensors"},{"rfilename":"model-00022-of-00050.safetensors"},{"rfilename":"model-00023-of-00050.safetensors"},{"rfilename":"model-00024-of-00050.safetensors"},{"rfilename":"model-00025-of-00050.safetensors"},{"rfilename":"model-00026-of-00050.safetensors"},{"rfilename":"model-00027-of-00050.safetensors"},{"rfilename":"model-00028-of-00050.safetensors"},{"rfilename":"model-00029-of-00050.safetensors"},{"rfilename":"model-00030-of-00050.safetensors"},{"rfilename":"model-00031-of-00050.safetensors"},{"rfilename":"model-00032-of-00050.safetensors"},{"rfilename":"model-00033-of-00050.safetensors"},{"rfilename":"model-00034-of-00050.safetensors"},{"rfilename":"model-00035-of-00050.safetensors"},{"rfilename":"model-00036-of-00050.safetensors"},{"rfilename":"model-00037-of-00050.safetensors"},{"rfilename":"model-00038-of-00050.safetensors"},{"rfilename":"model-00039-of-00050.safetensors"},{"rfilename":"model-00040-of-00050.safetensors"},{"rfilename":"model-00041-of-00050.safetensors"},{"rfilename":"model-00042-of-00050.safetensors"},{"rfilename":"model-00043-of-00050.safetensors"},{"rfilename":"model-00044-of-00050.safetensors"},{"rfilename":"model-00045-of-00050.safetensors"},{"rfilename":"model-00046-of-00050.safetensors"},{"rfilename":"model-00047-of-00050.safetensors"},{"rfilename":"model-00048-of-00050.safetensors"},{"rfilename":"model-00049-of-00050.safetensors"},{"rfilename":"model-00050-of-00050.safetensors"},{"rfilename":"model.safetensors.index.json"},{"rfilename":"modeling_nemotron_h.py"},{"rfilename":"privacy.md"},{"rfilename":"safety.md"},{"rfilename":"special_tokens_map.json"},{"rfilename":"super_v3_reasoning_parser.py"},{"rfilename":"tokenizer.json"},{"rfilename":"tokenizer_config.json"}],"spaces":["embedl/hfviewer","VIDraft/global-llm-leaderboard","hadadxyz/ai","FINAL-Bench/all-bench-leaderboard","yuan111556/ai","Nuanmanee/all-bench-leaderboard","Devil200501/ai","gdfgdgdfghghyt/ai","GGUFGuy/NVIDIA-Chatbot","hplisiecki/pinocchio-inventory-explorer"],"createdAt":"2026-03-10T18:32:14.000Z","safetensors":{"parameters":{"F32":20992,"BF16":123611012096},"total":123611012096},"inference":"warm","usedStorage":247244727964}