Image-Text-to-Text
Transformers
English
vision-language-model
vlm
surveillance
iot
gemma
vl-jepa
multimodal
object-detection
video-analytics
Instructions to use hardiksa/arcisvlm with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use hardiksa/arcisvlm with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("image-text-to-text", model="hardiksa/arcisvlm")# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("hardiksa/arcisvlm", dtype="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use hardiksa/arcisvlm with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "hardiksa/arcisvlm" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "hardiksa/arcisvlm", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/hardiksa/arcisvlm
- SGLang
How to use hardiksa/arcisvlm 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 "hardiksa/arcisvlm" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "hardiksa/arcisvlm", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'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 "hardiksa/arcisvlm" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "hardiksa/arcisvlm", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use hardiksa/arcisvlm with Docker Model Runner:
docker model run hf.co/hardiksa/arcisvlm
| # ========================================================================== | |
| # ArcisVLM Full Training Pipeline — 8x A100 DDP | |
| # | |
| # Runs all stages sequentially: | |
| # Stage 0: Train 32K BPE tokenizer | |
| # Stage 1: JEPA Alignment (~2.5 hrs) | |
| # Stage 2: Instruction Tuning (~1.5 hrs) | |
| # Stage 3: Domain Fine-Tuning (~30 min) | |
| # | |
| # Usage: | |
| # bash scripts/launch_training.sh | |
| # ========================================================================== | |
| set -e | |
| CONFIG=configs/scale_1.3b.yaml | |
| export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 | |
| export OMP_NUM_THREADS=4 | |
| export NCCL_IB_DISABLE=0 | |
| export NCCL_P2P_LEVEL=NVL | |
| # Change to project root | |
| cd "$(dirname "$0")/.." | |
| echo "==================================================================" | |
| echo "ArcisVLM Training Pipeline" | |
| echo "==================================================================" | |
| echo "Config: $CONFIG" | |
| echo "GPUs: $CUDA_VISIBLE_DEVICES" | |
| echo "Start time: $(date)" | |
| echo "==================================================================" | |
| # ------------------------------------------------------------------ | |
| echo "" | |
| echo "=== STAGE 0: TRAIN TOKENIZER ===" | |
| echo "Started: $(date)" | |
| python3 scripts/train_tokenizer_32k.py | |
| echo "Completed: $(date)" | |
| # ------------------------------------------------------------------ | |
| echo "" | |
| echo "=== STAGE 1: JEPA ALIGNMENT (~2.5 hrs) ===" | |
| echo "Started: $(date)" | |
| torchrun --nproc_per_node=8 scripts/train_stage1_ddp.py --config $CONFIG | |
| echo "Completed: $(date)" | |
| # ------------------------------------------------------------------ | |
| echo "" | |
| echo "=== STAGE 2: INSTRUCTION TUNING (~1.5 hrs) ===" | |
| echo "Started: $(date)" | |
| torchrun --nproc_per_node=8 scripts/train_stage2_ddp.py \ | |
| --config $CONFIG \ | |
| --stage1_ckpt checkpoints/stage1_final.pt | |
| echo "Completed: $(date)" | |
| # ------------------------------------------------------------------ | |
| echo "" | |
| echo "=== STAGE 3: DOMAIN FINE-TUNE (~30 min) ===" | |
| echo "Started: $(date)" | |
| torchrun --nproc_per_node=8 scripts/train_stage3_ddp.py \ | |
| --config $CONFIG \ | |
| --stage2_ckpt checkpoints/stage2_final.pt | |
| echo "Completed: $(date)" | |
| # ------------------------------------------------------------------ | |
| echo "" | |
| echo "=== PUSH CHECKPOINTS ===" | |
| python3 scripts/push_checkpoints.py || echo "[WARN] Push failed, manual push needed" | |
| # ------------------------------------------------------------------ | |
| echo "" | |
| echo "==================================================================" | |
| echo "TRAINING COMPLETE" | |
| echo "Finished: $(date)" | |
| echo "==================================================================" | |
| echo "" | |
| echo "Checkpoints:" | |
| ls -lh checkpoints/stage*_final.pt 2>/dev/null || echo " (none found)" | |
| echo "" | |
| echo "Next steps:" | |
| echo " 1. Run benchmarks: python3 scripts/run_benchmarks.py" | |
| echo " 2. Export for ONNX: python3 scripts/export_onnx.py" | |
| echo " 3. Deploy: python3 scripts/deploy.py" | |