--- license: other license_name: nvidia-open-model-license license_link: >- https://developer.download.nvidia.com/licenses/nvidia-open-model-license-agreement-june-2024.pdf inference: false fine-tuning: false tags: - fp8 - vllm base_model: mgoin/Nemotron-4-340B-Instruct-hf --- ## Nemotron-4-340B-Instruct-hf-FP8 Converted and quantized checkpoint of [nvidia/Nemotron-4-340B-Instruct](https://huggingface.co/nvidia/Nemotron-4-340B-Instruct). Specifically it was produced from the [v1.0 .nemo checkpoint on NGC](https://catalog.ngc.nvidia.com/orgs/nvidia/teams/nemo/models/nemotron-4-340b-instruct/files). This runs in vLLM with this PR: https://github.com/vllm-project/vllm/pull/6611 ### Evaluations All the below evaluations were run with `lm-eval==0.4.3` on 8xA100 GPUs. ``` lm_eval --model vllm --model_args pretrained=/home/mgoin/code/Nemotron-4-340B-Instruct-hf-FP8,tensor_parallel_size=8,distributed_executor_backend="ray",gpu_memory_utilization=0.6,enforce_eager=True --tasks mmlu --num_fewshot 0 --batch_size 4 vllm (pretrained=/home/mgoin/code/Nemotron-4-340B-Instruct-hf-FP8,tensor_parallel_size=8,distributed_executor_backend=ray,gpu_memory_utilization=0.6,enforce_eager=True), gen_kwargs: (None), limit: None, num_fewshot: 0, batch_size: 4 | Tasks |Version|Filter|n-shot|Metric| |Value | |Stderr| |---------------------------------------|-------|------|-----:|------|---|-----:|---|-----:| |mmlu |N/A |none | 0|acc |↑ |0.7905|± |0.0033| | - abstract_algebra | 0|none | 0|acc |↑ |0.5100|± |0.0502| | - anatomy | 0|none | 0|acc |↑ |0.8074|± |0.0341| | - astronomy | 0|none | 0|acc |↑ |0.8816|± |0.0263| | - business_ethics | 0|none | 0|acc |↑ |0.7300|± |0.0446| | - clinical_knowledge | 0|none | 0|acc |↑ |0.8453|± |0.0223| | - college_biology | 0|none | 0|acc |↑ |0.9236|± |0.0222| | - college_chemistry | 0|none | 0|acc |↑ |0.5300|± |0.0502| | - college_computer_science | 0|none | 0|acc |↑ |0.7400|± |0.0441| | - college_mathematics | 0|none | 0|acc |↑ |0.4500|± |0.0500| | - college_medicine | 0|none | 0|acc |↑ |0.7630|± |0.0324| | - college_physics | 0|none | 0|acc |↑ |0.5490|± |0.0495| | - computer_security | 0|none | 0|acc |↑ |0.8300|± |0.0378| | - conceptual_physics | 0|none | 0|acc |↑ |0.8043|± |0.0259| | - econometrics | 0|none | 0|acc |↑ |0.7105|± |0.0427| | - electrical_engineering | 0|none | 0|acc |↑ |0.7448|± |0.0363| | - elementary_mathematics | 0|none | 0|acc |↑ |0.6667|± |0.0243| | - formal_logic | 0|none | 0|acc |↑ |0.5952|± |0.0439| | - global_facts | 0|none | 0|acc |↑ |0.4900|± |0.0502| | - high_school_biology | 0|none | 0|acc |↑ |0.9097|± |0.0163| | - high_school_chemistry | 0|none | 0|acc |↑ |0.7143|± |0.0318| | - high_school_computer_science | 0|none | 0|acc |↑ |0.9100|± |0.0288| | - high_school_european_history | 0|none | 0|acc |↑ |0.8788|± |0.0255| | - high_school_geography | 0|none | 0|acc |↑ |0.9242|± |0.0189| | - high_school_government_and_politics| 0|none | 0|acc |↑ |0.9845|± |0.0089| | - high_school_macroeconomics | 0|none | 0|acc |↑ |0.8333|± |0.0189| | - high_school_mathematics | 0|none | 0|acc |↑ |0.4630|± |0.0304| | - high_school_microeconomics | 0|none | 0|acc |↑ |0.8824|± |0.0209| | - high_school_physics | 0|none | 0|acc |↑ |0.6159|± |0.0397| | - high_school_psychology | 0|none | 0|acc |↑ |0.9394|± |0.0102| | - high_school_statistics | 0|none | 0|acc |↑ |0.7639|± |0.0290| | - high_school_us_history | 0|none | 0|acc |↑ |0.9412|± |0.0165| | - high_school_world_history | 0|none | 0|acc |↑ |0.9409|± |0.0153| | - human_aging | 0|none | 0|acc |↑ |0.8072|± |0.0265| | - human_sexuality | 0|none | 0|acc |↑ |0.8855|± |0.0279| | - humanities |N/A |none | 0|acc |↑ |0.7594|± |0.0060| | - international_law | 0|none | 0|acc |↑ |0.9091|± |0.0262| | - jurisprudence | 0|none | 0|acc |↑ |0.8704|± |0.0325| | - logical_fallacies | 0|none | 0|acc |↑ |0.8528|± |0.0278| | - machine_learning | 0|none | 0|acc |↑ |0.6786|± |0.0443| | - management | 0|none | 0|acc |↑ |0.8641|± |0.0339| | - marketing | 0|none | 0|acc |↑ |0.9359|± |0.0160| | - medical_genetics | 0|none | 0|acc |↑ |0.8400|± |0.0368| | - miscellaneous | 0|none | 0|acc |↑ |0.9221|± |0.0096| | - moral_disputes | 0|none | 0|acc |↑ |0.8382|± |0.0198| | - moral_scenarios | 0|none | 0|acc |↑ |0.6168|± |0.0163| | - nutrition | 0|none | 0|acc |↑ |0.8791|± |0.0187| | - other |N/A |none | 0|acc |↑ |0.8214|± |0.0065| | - philosophy | 0|none | 0|acc |↑ |0.8521|± |0.0202| | - prehistory | 0|none | 0|acc |↑ |0.8796|± |0.0181| | - professional_accounting | 0|none | 0|acc |↑ |0.6383|± |0.0287| | - professional_law | 0|none | 0|acc |↑ |0.6838|± |0.0119| | - professional_medicine | 0|none | 0|acc |↑ |0.8824|± |0.0196| | - professional_psychology | 0|none | 0|acc |↑ |0.8611|± |0.0140| | - public_relations | 0|none | 0|acc |↑ |0.8000|± |0.0383| | - security_studies | 0|none | 0|acc |↑ |0.8204|± |0.0246| | - social_sciences |N/A |none | 0|acc |↑ |0.8811|± |0.0057| | - sociology | 0|none | 0|acc |↑ |0.9055|± |0.0207| | - stem |N/A |none | 0|acc |↑ |0.7180|± |0.0076| | - us_foreign_policy | 0|none | 0|acc |↑ |0.9600|± |0.0197| | - virology | 0|none | 0|acc |↑ |0.5482|± |0.0387| | - world_religions | 0|none | 0|acc |↑ |0.9006|± |0.0229| | Groups |Version|Filter|n-shot|Metric| |Value | |Stderr| |------------------|-------|------|-----:|------|---|-----:|---|-----:| |mmlu |N/A |none | 0|acc |↑ |0.7905|± |0.0033| | - humanities |N/A |none | 0|acc |↑ |0.7594|± |0.0060| | - other |N/A |none | 0|acc |↑ |0.8214|± |0.0065| | - social_sciences|N/A |none | 0|acc |↑ |0.8811|± |0.0057| | - stem |N/A |none | 0|acc |↑ |0.7180|± |0.0076| ``` The [original paper](https://d1qx31qr3h6wln.cloudfront.net/publications/Nemotron_4_340B_8T_0.pdf) evals for reference: ![image/png](https://cdn-uploads.huggingface.co/production/uploads/60466e4b4f40b01b66151416/R98q-Uh-3n-crwGcSLkuS.png)