mitra-qwen35-2b-base-stage2

The compact (2B) member of the Dharmamitra Qwen3.5 family: Qwen3.5-2B after ~45B tokens of continued pretraining on classical Buddhist corpora (2 epochs, 8k context) and a stage-2 SFT that adds translation-specific capabilities — bidirectional parallel translation and production translation turnarounds (1.4 epochs).

Use it where the 9B mitra-qwen35-base-stage2 is too heavy; same conventions (Tibetan in Wylie, Sanskrit/Pāli in IAST), same chat template.

Usage

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

tok = AutoTokenizer.from_pretrained("buddhist-nlp/mitra-qwen35-2b-base-stage2")
model = AutoModelForCausalLM.from_pretrained(
    "buddhist-nlp/mitra-qwen35-2b-base-stage2", dtype=torch.bfloat16, device_map="cuda"
)
messages = [{"role": "user", "content":
    "Translate into English: evaṃ mayā śrutam ekasmin samaye"}]
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=128)
print(tok.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Citation

If you use this model, please cite the Dharmamitra project (https://dharmamitra.org). A technical report is in preparation.