OmicAI-9B
The 9B member of OmicAI, a family optimized for omics data analysis and tool interaction over omicverse. Fine-tuned with ms-swift LoRA (all-linear, r16, α32) merged on a Qwen3.5-9B-architecture base (MiMo-V2.6-Distill-Qwen-9B).
The OmicAI family
| Model | Base | Access |
|---|---|---|
| OmicAI-0.8B | Qwen3.5-0.8B | Open — Apache-2.0 |
| OmicAI-4B | Qwen3.5-4B | Open — Apache-2.0 |
| OmicAI-9B (this model) | MiMo-V2.6-Distill-Qwen-9B | Open — Apache-2.0 |
| OmicAI-27B | Qwen3.8-27B | By application |
Training Data
Trajectories produced by OmicOS and OmicVerse through a paper-reproduction process over the last three years of publications (retrieve recent omics papers → reproduce their analyses end-to-end → multi-turn agent trajectories of retrieval, code, omicverse tool calls, execution-feedback recovery). Kept only when a judge confirms legitimacy and a scoreboard confirms the result matches or surpasses the paper's reported (SOTA) numbers.
Key Details
| Base model | XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B (Qwen3.5-9B architecture) |
| Method | ms-swift LoRA (all-linear, r16, α32) → merged |
| Chat template | qwen3_5 — inference template must match training |
| Precision | bfloat16 (4 shards) |
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("omicverse/OmicAI-9B")
model = AutoModelForCausalLM.from_pretrained("omicverse/OmicAI-9B", torch_dtype="bfloat16", device_map="auto")
Ecosystem
Part of omicOS / omicAI. Siblings: OmicAI-4B · OmicAI-27B · OmicAI-0.8B.