Nova Falcon H1R (Merged)
LoRA fine-tuned merge of tiiuae/Falcon-H1R-7B for the Nova Intelligence assistant.
Model Details
- Architecture: FalconH1ForCausalLM (hybrid Mamba2 + Attention)
- Parameters: 7.59B (BF16)
- Context Length: 262,144 tokens (256K)
- Base Model: tiiuae/Falcon-H1R-7B
- Training: LoRA fine-tuned and merged
Deployment
Recommended vLLM settings:
- Engine: vLLM >= 0.16.0
- Instance: Nvidia A100 80GB (1x)
- dtype: bfloat16 (MANDATORY - float16 not supported)
- max-model-len: 32768
- max-num-seqs: 64
- temperature: 0.1
- Downloads last month
- 4