sam860/
olmo2:1b-Q4_K_XL

14 1 month ago

Allen Institute's Latest tiny LLM in the Olmo family

1b

1 month ago

cde8cc651d36 · 967MB ·

model
olmo2
·
1.48B
·
Q4_K_M
template
{{- range $i, $_ := .Messages }} {{- $last := eq (len (slice $.Messages $i)) 1 -}} <|{{ .Role }}|> {

Readme

This 1B parameter instruct model is light enough for near‑instant local response while still competent for math and reasoning due to RLVR fine‑tuning.

OLMo 2 1B Instruct (Apr 2025) — compact, open, and tuned for reasoning-heavy prompts. Post-trained on Tülu 3 plus DPO and RLVR‑MATH, making it stronger on problem‑solving than most 1B models.
Ideal for:
- Lightweight chat + reasoning on CPU/low‑VRAM setups
- Math and structured Q&A (GSM8K, MATH)
- General instruct tasks where latency matters

References

HuggingFace