Instructions to use kenpath/mahavistaar-llm-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use kenpath/mahavistaar-llm-v1 with Transformers:

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

pipe = pipeline("text-generation", model="kenpath/mahavistaar-llm-v1")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("kenpath/mahavistaar-llm-v1")
model = AutoModelForMultimodalLM.from_pretrained("kenpath/mahavistaar-llm-v1")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.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(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use kenpath/mahavistaar-llm-v1 with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/kenpath/mahavistaar-llm-v1

SGLang

How to use kenpath/mahavistaar-llm-v1 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 "kenpath/mahavistaar-llm-v1" \
    --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": "kenpath/mahavistaar-llm-v1",
		"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 "kenpath/mahavistaar-llm-v1" \
        --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": "kenpath/mahavistaar-llm-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use kenpath/mahavistaar-llm-v1 with Docker Model Runner:
```
docker model run hf.co/kenpath/mahavistaar-llm-v1
```

mhv_synthetic_qwen3.5_v0.3

Fine-tuned version of kenpath/mhv_mhv-_all_qwen3.5-27b_v0.3 for the Vistaar (unknown) use case.

Model Details

Base Model: kenpath/mhv_mhv-_all_qwen3.5-27b_v0.3
Fine-tuning Method: LoRA (Low-Rank Adaptation)
Training Framework: HuggingFace Transformers + PEFT + DeepSpeed ZeRO-3
Merged: Full model weights (LoRA merged into base)

Training Configuration

LoRA Parameters

Parameter	Value
Rank (r)	64
Alpha	64
Dropout	0.1
Target Modules	['q_proj', 'k_proj', 'v_proj', 'o_proj', 'gate_proj', 'up_proj', 'down_proj']
Trainable Parameters	318,767,104 / 27,214,765,568 (1.1713%)

Training Hyperparameters

Parameter	Value
Epochs	1
Per-device Batch Size	1
Gradient Accumulation	1
Effective Batch Size	1
Learning Rate	5e-06
LR Scheduler	cosine
Warmup Ratio	N/A
Weight Decay	0.001
Max Grad Norm	1.0

Hardware & Performance

Metric	Value
GPUs	8x H100 80GB
Training Duration	5.89 hours
Final Training Loss	0.4769
Dataset Size	2,226 samples
Max Sequence Length	100000

W&B Run: https://wandb.ai/kenpath-ai-team/mh-finetuning/runs/rde8a76q

Dataset

Name: combined-mh-synthetic-vistaar
Samples Used: N/A

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "kenpath/mhv_synthetic_qwen3.5_v0.3",
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained("kenpath/mhv_synthetic_qwen3.5_v0.3")

messages = [
    {"role": "user", "content": "Your question here"}
]

input_ids = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt",
).to(model.device)

output = model.generate(input_ids, max_new_tokens=512)
print(tokenizer.decode(output[0], skip_special_tokens=True))

License

This model inherits the license from the base model kenpath/mhv_mhv-_all_qwen3.5-27b_v0.3.

Generated on 2026-03-16 05:21:25 | Merge took 4.8s

Downloads last month: 322

Safetensors

Model size

27B params

Tensor type

BF16

Model tree for kenpath/mahavistaar-llm-v1

Base model

Qwen/Qwen3.5-27B

Adapter

kenpath/mhv_mhv-_all_qwen3.5-27b_v0.3

Adapter

(1)

this model