NVIDIA
Llama-3.1-Nemotron-Ultra-253B-v1
Container
NVIDIA
Llama-3.1-Nemotron-Ultra-253B-v1

This container houses the Llama-3.1-Nemotron-Ultra-253B-v1, a reasoning model offering a great tradeoff between accuracy and efficiency. Post-trained for chat, RAG, and tool calling, it supports a 128K context and fits on a single 8xH100 node.

LayerLabelCreated
sha256:a3ed95caeb02ffe68cdd9fd84406680ae93d633cb16422d00e8a7c22955b46d4USER
1000
05/29/2025 1:31 AM UTC
sha256:943e5f36050543e9692ff2061ef9b45ff3fcdd6d91004f243dc4e6f3c19c0755RUN
NIM_MODEL_NAME=nvidia/llama-3.1-nemotron-ultra-253b-v1 NIM_VERSION=1.8.4 NIM_CONTAINER_NSPECT_ID=NSPECT-KNVB-G50E NIM_MODEL_NSPECT_ID=NSPECT-GQ1G-FQ7S NIM_TYPE=llm NIM_USER_ID=1000 rm /opt/nvidia/entrypoint.d/30-container-license.txt &&
  mkdir /.triton &&
  chown ${NIM_USER_ID}:${NIM_USER_ID} /.triton &&
  mkdir -p /etc/nim/config/ &&
  ln -s /opt/nim/etc/default/model_manifest.yaml /etc/nim/config/model_manifest.yaml &&
  mkdir /llama-nemotron-ultra &&
  python3 -c 'from nim_sdk.hub import FileSet, WorkspaceComponent, Workspace; fs = FileSet("ngc://nvstaging/nim/llama-3.1-nemotron-ultra-253b-v1:hf-e155ab2-nim-bf16-tool-use", globs=["*.json", "*.py"]); wsc = WorkspaceComponent(fs, ""); ws = Workspace([wsc]); ws = ws.make_fully_defined(); repo = ws.repo(); files = repo.get_all(); files.copy_all("/llama-nemotron-ultra")' &&
  mkdir /opt/nim/llm/nim_llm_sdk/model_specific_modules/llama-nemotron-ultra &&
  cp /llama-nemotron-ultra/*.json /opt/nim/llm/nim_llm_sdk/model_specific_modules/llama-nemotron-ultra &&
  cp /llama-nemotron-ultra/*.py /opt/nim/llm/nim_llm_sdk/model_specific_modules/llama-nemotron-ultra &&
  rm -rf /llama-nemotron-ultra &&
  chown -R ${NIM_USER_ID}:${NIM_USER_ID} /opt/nim/llm/nim_llm_sdk/model_specific_modules/llama-nemotron-ultra &&
  if [ -f /opt/nim/start-server.sh ]; then ln -s /opt/nim/start-server.sh /opt/nim/start_server.sh; fi &&
  sed -i 's/python3 -m nim_llm_sdk.entrypoints.launch/python3 -m nim_llm_sdk.entrypoints.launch --enable-auto-tool-choice --tool-call-parser llama3_json/g' /opt/nim/start_server.sh &&
  MISTRAL_FILE=$(find /opt/nim -name "mistralv3.py" 2>/dev/null | grep -m 1 "guided_decoding/models") &&
  if [ -n "$MISTRAL_FILE" ]; then echo "Found mistralv3.py at $MISTRAL_FILE" &&
  sed -i '/def check_tools_state/,/return ToolUseState.UNUSED/{/def check_tools_state/!{/return ToolUseState.UNUSED/!d}}' "$MISTRAL_FILE" &&
  sed -i '/def check_tools_state/a\        model_output = model_output.strip()\n        if model_output.startswith(self.bot_token):\n            return ToolUseState.USED\n        elif model_output.startswith("<"):\n            return ToolUseState.POSSIBLE\n        else:\n            return ToolUseState.UNUSED' "$MISTRAL_FILE" &&
  echo "Successfully patched $MISTRAL_FILE"; else echo "WARNING: Could not find mistralv3.py file" &&
  find /opt -name "mistralv3.py" 2>/dev/null ||
  echo "No mistralv3.py files found in /opt"; fi
05/29/2025 1:31 AM UTC
sha256:a3ed95caeb02ffe68cdd9fd84406680ae93d633cb16422d00e8a7c22955b46d4USER
0
05/29/2025 1:30 AM UTC
sha256:5ef4b3244f05ffed82c238ece339bfec79f18400436210b03fbadc67493a5e35COPY
CONTAINER_LICENSE /opt/nim/LICENSE
05/29/2025 1:30 AM UTC
sha256:fdabaa11a31bb92970e0ca3f9cff703c26e7efd632bd137bb0dd5f5bfbaa0364COPY
./notices/* /opt/nim/
05/29/2025 1:30 AM UTC
sha256:0d9ad3fd33b4458bc1c65391c8c81bddb388cb078d43415658d4084d0945bc27COPY
banner.txt /opt/nvidia/entrypoint.d/31-model-license.txt
05/29/2025 1:30 AM UTC
sha256:4abd27c7c85ac66fce842272abc976ff5a37078ec38ea5fa0c66adaeec16f63bCOPY
MODEL_LICENSE /opt/nim/MODEL_LICENSE
05/29/2025 1:30 AM UTC
sha256:9d982cbddc6182cdaaf77632654747bc7c6e8d25bc0404882ae0a00277e3b06aCOPY
model_manifest.yaml /opt/nim/etc/default/model_manifest.yaml
05/29/2025 1:30 AM UTC
sha256:a3ed95caeb02ffe68cdd9fd84406680ae93d633cb16422d00e8a7c22955b46d4ENV
NIM_HTTP_API_PORT=8000
05/29/2025 1:30 AM UTC
sha256:a3ed95caeb02ffe68cdd9fd84406680ae93d633cb16422d00e8a7c22955b46d4ENV
INFERENCE_MICROSERVICE_LLM_NIM_VERSION=1.8.4
05/29/2025 1:30 AM UTC
...

NVIDIA uses cookies to improve your experience on our web site. We and our third-party partners also use cookies and other tools to collect and record information you provide as well as information about your interactions with our websites for performance improvement, analytics, and to assist in marketing efforts. By clicking "Accept All", you consent to our use of cookies and other tools as described in our Cookie Policy. You can manage your cookie settings by clicking on "Manage Settings." By continuing to use this site or by clicking one of the buttons below, you agree to our Terms of Service (which contains important waivers). Please see our Privacy Policy for more information on our privacy practices.