mitra-qwen35-it
The general-purpose instruction/chat model of the
Dharmamitra Qwen3.5 family: a multi-turn SFT of
mitra-qwen35-base-stage2
for Buddhist-studies conversation — closed-book Buddhism Q&A plus
translation and translation-refinement assistance for classical languages
(Sanskrit, Tibetan, Buddhist Chinese, Pāli).
This is the recommended entry point if you want to talk to the mitra family; use the stage-2 base for raw translation pipelines and the embedder for retrieval.
Usage
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("buddhist-nlp/mitra-qwen35-it")
model = AutoModelForCausalLM.from_pretrained(
"buddhist-nlp/mitra-qwen35-it", dtype=torch.bfloat16, device_map="cuda"
)
messages = [{"role": "user", "content":
"What is the difference between śamatha and vipaśyanā?"}]
inputs = tok.apply_chat_template(messages, add_generation_prompt=True,
return_dict=True, return_tensors="pt").to(model.device)
out = model.generate(**inputs, max_new_tokens=512)
print(tok.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))
Tibetan may be given in Wylie transliteration; Sanskrit and Pāli in IAST.
Training details
- Base:
buddhist-nlp/mitra-qwen35-base-stage2(9B; ~30B tokens Buddhist CPT- translation-capability SFT)
- Multi-turn SFT (TRL, completion-only loss over full chat histories), LR 1e-5, on a combined corpus of closed-book Buddhism Q&A (knowledge distilled into weights) and translation/refinement tasks across zh/sa/bo/pi
Citation
If you use this model, please cite the Dharmamitra project (https://dharmamitra.org). A technical report is in preparation.