CoolFace
Modelpublic

sugiv/qwen3-8b-tanglish

sourceHugging Faceapache-2.0updated 2mo agoView on Hugging Face
0likes95downloads
Model Card

Qwen3-8B · Tanglish LoRA v1

Fine-tune of `Qwen/Qwen3-8B` that replies in casual Tanglish — code-mixed Tamil transliterated into the Latin alphabet, as spoken every day in Chennai and across South India.

Trained on sugiv/tanglish-pairs-v1 (81,261 SFT examples) with LoRA r=16, alpha=32 on the bf16 base (not QLoRA/4-bit — L40S 48 GB has enough VRAM for cleaner training). This repo ships the fully merged bf16 model (16.4 GB) plus a standalone PEFT adapter and all 27 intermediate resume checkpoints.

Highlights (from Phase 4 eval)

  • —Beats stock Qwen3-8B on every LLM-judge dimension on both single-turn and multi-turn Tanglish prompts. Full report at `eval/PHASE_4_EVAL.md`.
  • —Judge: qwen3-235b-a22b-instruct-2507, temperature=0, 40/40 parses successful.
Dimension (1-5, higher = better)tngbasedelta
single-turn intelligibility4.874.20+0.67
single-turn tanglish_authenticity4.333.00+1.33
single-turn helpfulness4.003.33+0.67
single-turn naturalness4.603.53+1.07
multi-turn intelligibility5.004.80+0.20
multi-turn tanglish_authenticity4.003.60+0.40
multi-turn helpfulness4.603.60+1.00
multi-turn naturalness4.804.00+0.80
  • —7.5x faster inference than base: 1.91 s vs 14.39 s mean single-turn on L4. Base emits a <think>Okay, the user is asking...</think> reasoning dump on every casual chat; this fine-tune was trained on outputs that keep <think></think> empty so it goes straight to the Tanglish reply.
  • —Zero Tamil-script leaks on 15 held-out prompts (base has 3/15).

Training (verified from training/qwen_train.log + trainer_state.json)

SettingValue
Base modelQwen/Qwen3-8B (bf16, flashattention2, gradient_checkpointing)
AdapterLoRA r=16, alpha=32, target = all-linear
Precisionbf16 (not QLoRA / 4-bit)
Corpussugiv/tanglish-pairs-v1, 77,198 train + 4,063 val
Effective batch size32 (per-device 4 x grad-accum 8)
Learning rate2e-4 cosine, warmup steps = 100
Total training steps~14,475 (3 full epochs, last saved checkpoint at 13,500)
Best `eval_loss`0.8126 at step 13,500
First eval (step 500)1.1184
Eval-loss trajectorymonotonically decreasing across all 27 checkpoints
OptimizerAdamW
Hardware1x L40S 48 GB SECURE (RunPod, US-KS-1)
Wall time~11 hours
Cost~$13.36

Full step-by-step training log at `training/qwen_train.log` (2.1 MB, 702 log entries). Config at `training/qwen_train_tanglish.yaml`.

Usage

Merged model (recommended, no PEFT install needed)

python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "sugiv/qwen3-8b-tanglish",
    torch_dtype="bfloat16",
    device_map="cuda",
    token="hf_...",
)
tok = AutoTokenizer.from_pretrained("sugiv/qwen3-8b-tanglish", token="hf_...")

msgs = [
    {"role": "system", "content": "You are a friendly Tanglish-speaking assistant. Reply naturally in casual, code-mixed Tanglish..."},
    {"role": "user", "content": "machi, nalaikku Chennai la enna weather?"},
]
prompt = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt").to("cuda")
out = model.generate(prompt, max_new_tokens=200, temperature=0.7, top_p=0.9, do_sample=True)
print(tok.decode(out[0][prompt.shape[-1]:], skip_special_tokens=True))
# => 'da, innum hot ah iruku, morning la 28 degree nu solraanga.'

LoRA adapter (attach to a stock Qwen3-8B)

python
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-8B", torch_dtype="bfloat16", device_map="cuda")
model = PeftModel.from_pretrained(base, "sugiv/qwen3-8b-tanglish", subfolder="lora", token="hf_...")

vLLM / RunPod Serverless (production)

bash
# On Runpod Hub: pick `runpod-workers/worker-vllm`, set:
#   MODEL_NAME     = sugiv/qwen3-8b-tanglish
#   DTYPE          = bfloat16
#   MAX_MODEL_LEN  = 2048
#   HF_TOKEN       = hf_...
# GPU: L40S 48 GB SECURE recommended (fits with headroom).

Resume from any of the 27 intermediate checkpoints

Every checkpoint under checkpoints/checkpoint-{500,1000,...,13500}/ contains the LoRA adapter, optimizer state, LR scheduler state, RNG state, and trainer_state.json — enough to resume TRL SFTTrainer from that exact step.

Repo contents

PathBytesWhat
model-000{1..4}-of-00004.safetensors~16.4 GBMerged bf16 weights (base + LoRA collapsed)
config.json + tokenizer* + chat_template.jinja~14 MBSame as base Qwen3-8B, unmodified
lora/adapter_model.safetensors~175 MBLoRA-only best adapter (step 13500)
lora/{adapter_config,training_args,tokenizer*}~14 MBPEFT metadata + tokenizer
checkpoints/checkpoint-*/~14 GB (27 dirs)Resume checkpoints every 500 steps, includes optimizer.pt + scheduler.pt
eval/MANIFEST.json~80 KB40 records with LLM-judge scores across 4 dimensions
eval/PHASE_4_EVAL.md~6 KBHuman-readable eval report with side-by-side samples
training/qwen_train.log~2 MBFull stdout from the 11-hour training run
training/*.yaml~5 KBTraining config used

Known limitations

  • —Soft-refusal on safety: the model deflects "how do I hack a database" with playful Tanglish banter instead of a textbook refusal. Judge deducts helpfulness=1 but flags the response as safe. If you need a stricter tone, layer a system-prompt refusal template at the agent layer.
  • —Verbose reasoning mode is trained out: the base Qwen3-8B emits <think>...</think> reasoning blocks on casual chat, and this fine-tune suppresses that behaviour. If you want explicit reasoning, use the base model — this one goes straight to the answer.
  • —Language support: Tanglish (Latin-script Tamil + English code-mix) is the trained target. It can still reply in pure Tamil script or pure English if prompted, but the training corpus is 100% transliterated Tanglish.

License

Apache-2.0 (inherited from Qwen/Qwen3-8B). Commercial use allowed. Attribution to both this repo and the base model is appreciated.

Citation

bibtex
@misc{tanglish_qwen3_2026,
  title={Qwen3-8B Tanglish LoRA},
  author={sugiv},
  year={2026},
  url={https://huggingface.co/sugiv/qwen3-8b-tanglish}
}

Related