README.md

metadata

license: other
license_name: qwen
license_link: https://huggingface.co/Qwen/Qwen2.5-7B-Instruct/blob/main/LICENSE
language:
  - hi
  - en
base_model: Qwen/Qwen2.5-7B-Instruct
library_name: peft
pipeline_tag: text-generation
tags:
  - hinglish
  - hindi
  - code-mixing
  - qlora
  - lora
  - sft
  - unsloth
  - trl
  - conversational
  - qwen2
datasets:
  - HydraLM/hindi_english_romanized
  - NebulaByte/alpaca-gpt4-hindi-hinglish
  - findnitai/english-to-hinglish
model-index:
  - name: hinglish-qwen2.5-7b-instruct
    results:
      - task:
          type: text-generation
        metrics:
          - name: Eval Loss
            type: loss
            value: 0.899
          - name: Training Loss (final)
            type: loss
            value: 0.25
widget:
  - text: >
      <|im_start|>system

      Aap ek helpful AI assistant hain jo Hinglish mein baat kar sakta
      hai.<|im_end|>

      <|im_start|>user

      Yaar, mujhe Python seekhni hai. Kahan se start karun?<|im_end|>

      <|im_start|>assistant
    example_title: Learn Python (Hinglish)
  - text: >
      <|im_start|>system

      Aap ek helpful AI assistant hain jo Hinglish mein baat kar sakta
      hai.<|im_end|>

      <|im_start|>user

      Ghar pe pizza banana sikhao<|im_end|>

      <|im_start|>assistant
    example_title: Homemade Pizza (Hinglish)
  - text: >
      <|im_start|>system

      Aap ek helpful AI assistant hain jo Hinglish mein baat kar sakta
      hai.<|im_end|>

      <|im_start|>user

      Machine learning aur AI mein kya difference hai?<|im_end|>

      <|im_start|>assistant
    example_title: ML vs AI (Hinglish)

🇮🇳 Hinglish Qwen2.5-7B Instruct — LoRA Fine-Tune

A bilingual (Hindi-English / Hinglish) fine-tune of Qwen2.5-7B-Instruct using QLoRA + Unsloth, designed to fluently converse in Hinglish — the natural code-mixed language spoken by 350M+ people across India.

⚡ Key Results

Metric	Baseline	Fine-tuned	Change
Eval Loss	2.57	0.90	↓ 65%
Training Loss	1.63	0.25	—
Hinglish Response Rate	10%	80%	↑ 8×

---

🔍 Before vs After

	Baseline (Qwen2.5-7B)	Fine-tuned (Ours)
"Yaar, Python seekhni hai"	Pure English bullet points, broken numbering	Natural Hinglish advice like a friend would give
"Ghar pe pizza banana sikhao"	Textbook English recipe format	Casual Hinglish with Indian context ("Domino's se bhi better!")
"Stock market invest karna chahiye?"	Generic English financial advice	Practical Hinglish with real talk ("Loan lekar mat karna!")

---

Model Details

Field	Value
Base Model	Qwen2.5-7B-Instruct (4-bit NF4)
Method	QLoRA — LoRA adapters on quantized base
Framework	Unsloth + HuggingFace TRL SFTTrainer
Hardware	NVIDIA RTX 3060 (12GB VRAM)
Precision	BF16 mixed precision
LoRA Config	r=32, α=64, RSLoRA, 7 target modules
Trainable Params	80.7M / 7.7B (1.05%)
Training Time	~2 hours, 675 steps, 3 epochs

---

Training

Parameter	Value
Epochs	3 (675 total steps)
Effective Batch Size	16 (1 × 16 grad accum)
Learning Rate	2e-4 → cosine decay
Warmup	50 steps
Optimizer	AdamW 8-bit
Eval Strategy	Every 100 steps
Best Checkpoint	Step 400 (eval loss 0.899)

Loss Curve

---

Dataset

4,000 curated samples from three sources:

Source	Samples
HuggingFace (HydraLM, NebulaByte, findnitai)	~2,000
Gemini Synthetic (5 categories)	~1,000
Alpaca → Hinglish Translation	~1,000

Format: ChatML with Hinglish system prompt. Split: 90% train / 10% eval.

---

Usage

from unsloth import FastLanguageModel
from peft import PeftModel

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="unsloth/Qwen2.5-7B-Instruct-bnb-4bit",
    max_seq_length=2048,
    load_in_4bit=True,
)
model = PeftModel.from_pretrained(model, "psyka-101/hinglish-qwen2.5-7b-instruct")
FastLanguageModel.for_inference(model)

messages = [
    {"role": "system", "content": "Aap ek helpful AI assistant hain jo Hinglish mein baat kar sakta hai."},
    {"role": "user", "content": "Yaar, mujhe Python seekhni hai. Kahan se start karun?"},
]
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt").to("cuda")
outputs = model.generate(inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

---

Limitations

• Trained on ~4K samples — may not generalize to all dialects or specialized topics
• Checkpoint-400 recommended (eval loss rises in epoch 3)
• May still default to pure English on some prompts
• Roman-script Hinglish only — no Devanagari
• Not tested for safety/toxicity

License

Inherits the Qwen2.5 license.

---

_{Built with 🦥 Unsloth • 🤗 Transformers • 🇮🇳 Hinglish}