Instructions to use manishiitg/open-aditi-hi-v2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use manishiitg/open-aditi-hi-v2 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="manishiitg/open-aditi-hi-v2")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("manishiitg/open-aditi-hi-v2")
model = AutoModelForCausalLM.from_pretrained("manishiitg/open-aditi-hi-v2", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use manishiitg/open-aditi-hi-v2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "manishiitg/open-aditi-hi-v2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "manishiitg/open-aditi-hi-v2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/manishiitg/open-aditi-hi-v2

SGLang

How to use manishiitg/open-aditi-hi-v2 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "manishiitg/open-aditi-hi-v2" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "manishiitg/open-aditi-hi-v2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "manishiitg/open-aditi-hi-v2" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "manishiitg/open-aditi-hi-v2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use manishiitg/open-aditi-hi-v2 with Docker Model Runner:
```
docker model run hf.co/manishiitg/open-aditi-hi-v2
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Model trained on Hindi and English data.

Try it out: https://colab.research.google.com/drive/1A_hbsq1vrCeAh3dEMvtwxxNxcNZ1BUyW?usp=sharing

For sample responose on different prompts checkout: https://github.com/manishiitg/hi-llm-eval

Language Hi

Model	implicit_hate	flores	indicwikibio	hellaswag-indic	truthfulqa-hi	boolq-hi	indicheadline	indic-arc-easy	indicqa	indic-arc-challenge	indicsentiment	xlsum-hi	indicxparaphrase	mmlu_hi
open-aditi-hi-v2	11.5021	43.6822	0.4846	0.2404	0.6934	0.8541	0.4565	0.4979	0.0795	0.4462	0.9729	0.4213	0.6838	0.3253
OpenHermes-2.5-Mistral-7B	0.2068	30.3465	0.3332	0.2485	0.3234	0.5979	0.1996	0.3523	0.2721	0.3396	0.9048	0.1774	0.8766	0.2769
open-aditi-hi-v1	8.6105	40.2376	0.4104	0.0848	0.4230	0.3758	0.4248	0.3889	0.1306	0.3558	0.8798	0.4212	0.5939	0.1398
Airavata	0.0663	58.0555	0.0637	0.0254	0.2122	0.0373	0.4346	0.1128	0.1008	0.0836	0.8437	0.4650	0.3277	0.1336

Language En

Model	boolq	hellaswag	mmlu	truthfulqa	xlsum	arc-easy-exact	arc-challenge
OpenHermes-2.5-Mistral-7B	0.4061	0.7999	0.5991	0.2081	0.4328	0.8687	0.7790
open-aditi-hi-v2	0.3982	0.4738	0.5544	0.2999	0.4349	0.8388	0.7235
open-aditi-hi-v1	0.0434	0.3509	0.2597	0.3317	0.4288	0.7588	0.6271
Airavata	0.0437	0.0277	0.1165	0.3586	0.4393	0.2534	0.1630

Task: flores Metric: chrf

Task: implicit_hate Metric: chrf

Task: indicsentiment Metric: accuracy

Task: indicxparaphrase Metric: accuracy

Task: boolq-hi Metric: accuracy

Task: truthfulqa-hi Metric: accuracy

Task: indic-arc-easy Metric: accuracy

Task: indicwikibio Metric: bleurt

Task: xlsum-hi Metric: bleurt

Task: indicheadline Metric: bleurt

Task: indic-arc-challenge Metric: accuracy

Task: mmlu_hi Metric: average_acc

Task: indicqa Metric: accuracy

Task: hellaswag-indic Metric: accuracy

Task: arc-easy-exact Metric: accuracy

Task: hellaswag Metric: accuracy

Task: arc-challenge Metric: accuracy

Task: mmlu Metric: average_acc

Task: xlsum Metric: bleurt

Task: boolq Metric: accuracy

Task: truthfulqa Metric: accuracy

Model evaluation on OpenLLM LeaderBoard

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	59.31
AI2 Reasoning Challenge (25-Shot)	59.39
HellaSwag (10-Shot)	82.01
MMLU (5-Shot)	61.41
TruthfulQA (0-shot)	45.84
Winogrande (5-shot)	77.19
GSM8k (5-shot)	30.02

Downloads last month: 7

Safetensors

Model size

7B params

Tensor type

BF16

Model tree for manishiitg/open-aditi-hi-v2

Quantizations

2 models

Collection including manishiitg/open-aditi-hi-v2

Aditi LLM

Collection

LLM's finetune on hinglish/hindi and english • 9 items • Updated Jul 13, 2024 • 1

Evaluation results

normalized accuracy on AI2 Reasoning Challenge (25-Shot)
test set Open LLM Leaderboard

59.390
normalized accuracy on HellaSwag (10-Shot)
validation set Open LLM Leaderboard

82.010
accuracy on MMLU (5-Shot)
test set Open LLM Leaderboard

61.410
mc2 on TruthfulQA (0-shot)
validation set Open LLM Leaderboard

45.840
accuracy on Winogrande (5-shot)
validation set Open LLM Leaderboard

77.190
accuracy on GSM8k (5-shot)
test set Open LLM Leaderboard

30.020