README.md

---
license: other
license_name: gemma-terms-of-use
license_link: https://ai.google.dev/gemma/terms
base_model: google/gemma-2b
datasets:
- ravithejads/samvaad-hi-filtered
- Telugu-LLM-Labs/telugu_teknium_GPTeacher_general_instruct_filtered_romanized
- Telugu-LLM-Labs/telugu_alpaca_yahma_cleaned_filtered_romanized
- abhinand/tamil-alpaca
- Tensoic/airoboros-3.2_kn
- Tensoic/gpt-teacher_kn
- VishnuPJ/Alpaca_Instruct_Malayalam
- Tensoic/Alpaca-Gujarati
- HydraIndicLM/punjabi_alpaca_52K
- HydraIndicLM/bengali_alpaca_dolly_67k
- OdiaGenAI/Odia_Alpaca_instructions_52k
- yahma/alpaca-cleaned
language:
- te
- en
- ta
- ml
- hi
- kn
- gu
- bn
- pa
- or
library_name: transformers
pipeline_tag: text-generation
---

# Indic-gemma-2b-finetuned-sft-Navarasa

This model is based on [google/gemma-2b](https://huggingface.co/google/gemma-7b) and hase been LoRA finetuned on 9 Indian languages and English language instruction datasets:

1. #### Hindi - [ravithejads/samvaad-hi-filtered](https://huggingface.co/datasets/ravithejads/samvaad-hi-filtered), [HydraIndicLM/hindi_alpaca_dolly_67k](https://huggingface.co/datasets/HydraIndicLM/hindi_alpaca_dolly_67k)(sampled)
2. #### Telugu - [Telugu-LLM-Labs/yahma_alpaca_cleaned_telugu_filtered_and_romanized](https://huggingface.co/datasets/Telugu-LLM-Labs/yahma_alpaca_cleaned_telugu_filtered_and_romanized), [Telugu-LLM-Labs/teknium_GPTeacher_general_instruct_telugu_filtered_and_romanized](https://huggingface.co/datasets/Telugu-LLM-Labs/teknium_GPTeacher_general_instruct_telugu_filtered_and_romanized)
3. #### Tamil - [abhinand/tamil-alpaca](https://huggingface.co/datasets/abhinand/tamil-alpaca)
4. #### Kannada - [Tensoic/airoboros-3.2_kn](https://huggingface.co/datasets/Tensoic/airoboros-3.2_kn), [Tensoic/gpt-teacher_kn](https://huggingface.co/datasets/Tensoic/gpt-teacher_kn)
5. #### Malayalam - [VishnuPJ/Alpaca_Instruct_Malayalam](https://huggingface.co/datasets/VishnuPJ/Alpaca_Instruct_Malayalam)
6. #### Gujarati - [Tensoic/Alpaca-Gujarati](https://huggingface.co/datasets/Tensoic/Alpaca-Gujarati)
7. #### Punjabi - [HydraIndicLM/punjabi_alpaca_52K](https://huggingface.co/datasets/HydraIndicLM/punjabi_alpaca_52K)
8. #### Bengali - [HydraIndicLM/bengali_alpaca_dolly_67k](https://huggingface.co/datasets/HydraIndicLM/bengali_alpaca_dolly_67k)(alpaca filtered)
9. #### Odia - [OdiaGenAI/Odia_Alpaca_instructions_52k](https://huggingface.co/datasets/OdiaGenAI/Odia_Alpaca_instructions_52k), [OdiaGenAI/gpt-teacher-roleplay-odia-3k](https://huggingface.co/datasets/OdiaGenAI/gpt-teacher-roleplay-odia-3k)
10. #### English - [yahma/alpaca-cleaned](https://huggingface.co/datasets/yahma/alpaca-cleaned)

The model is finetuned using [unsloth](https://github.com/unslothai/unsloth) library and we provide inference code using the same for faster inference. Alternatively you can use HuggingFace Library for inference.

# Training Details:

The model is trained on approx 500K instruction samples.
1. GPU: 1 A100, 80GB
2. Time: 36.5 Hours
3. Platform: [E2E Networks](https://www.e2enetworks.com/)
# Installation

`!pip install "unsloth[colab-ampere] @git+https://github.com/unslothai/unsloth.git"`

# Input Text Format

```
### Instruction: {instruction}

### Input: {input}

## Response: {response}
```

# Inference With Unsloth

```python3
from unsloth import FastLanguageModel
import torch
max_seq_length = 2048
dtype = None # None for auto detection. Float16 for Tesla T4, V100, Bfloat16 for Ampere+
load_in_4bit = False 
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "Telugu-LLM-Labs/Indic-gemma-2b-finetuned-sft-Navarasa",
    max_seq_length = max_seq_length,
    dtype = dtype,
    load_in_4bit = load_in_4bit,
    device_map="auto"
)
FastLanguageModel.for_inference(model) # Enable native 2x faster inference

input_prompt = """
### Instruction:
{}

### Input:
{}

### Response:
{}"""

input_text = input_prompt.format(
        "Tranlsate following sentence to Hindi.", # instruction
        "This model is developed by Telugu LLM Labs", # input
        "", # output - leave this blank for generation!
    )

inputs = tokenizer([input_text], return_tensors = "pt").to("cuda")

outputs = model.generate(**inputs, max_new_tokens = 300, use_cache = True)
response = tokenizer.batch_decode(outputs)
```

# Inference with HuggingFace

```python3
from peft import AutoPeftModelForCausalLM
from transformers import AutoTokenizer

model = AutoPeftModelForCausalLM.from_pretrained(
    "Telugu-LLM-Labs/Indic-gemma-2b-finetuned-sft-Navarasa",
    load_in_4bit = False,
    token = hf_token
)
tokenizer = AutoTokenizer.from_pretrained("Telugu-LLM-Labs/Indic-gemma-2b-finetuned-sft-Navarasa")

input_prompt = """
### Instruction:
{}

### Input:
{}

### Response:
{}"""

input_text = input_prompt.format(
        "Tranlsate following sentence to Hindi.", # instruction
        "This model is developed by Telugu LLM Labs", # input
        "", # output - leave this blank for generation!
    )

inputs = tokenizer([input_text], return_tensors = "pt").to("cuda")

outputs = model.generate(**inputs, max_new_tokens = 300, use_cache = True)
response = tokenizer.batch_decode(outputs)[0]
```

Refer to the [blog post](https://ravidesetty.medium.com/introducing-indic-gemma-7b-2b-instruction-tuned-model-on-9-indian-languages-navarasa-86bc81b4a282) for sample examples.

Please check our [Code Repository](https://github.com/TeluguLLMLabs/Indic-gemma-7b-Navarasa)for training and inference scripts.

# Developers:

The model is a collaborative effort by [Ravi Theja](https://twitter.com/ravithejads) and [Ramsri Goutham](https://twitter.com/ramsri_goutham). Feel free to DM either of us if you have any questions.