
Gemini
Google's family of multimodal AI models.

A 15B parameter model for code generation trained on 600+ programming languages.

StarCoder2-15B is a 15 billion parameter language model designed for code generation. It is trained on over 600 programming languages from The Stack v2 dataset, excluding opt-out requests. The model uses Grouped Query Attention and has a context window of 16,384 tokens with a sliding window attention of 4,096 tokens. Trained with the Fill-in-the-Middle objective on 4+ trillion tokens using NVIDIA NeMo framework on NVIDIA DGX H100 systems. It's not an instruction-following model, it excels at generating code snippets given context. Fine-tuning scripts are available in the StarCoder2's GitHub repository. Quantized versions are available through bitsandbytes for efficient memory usage.
StarCoder2-15B is a 15 billion parameter language model designed for code generation.
Explore all tools that specialize in generate code. This domain focus ensures StarCoder2-15B delivers optimized results for this specific requirement.
Explore all tools that specialize in complete code. This domain focus ensures StarCoder2-15B delivers optimized results for this specific requirement.
Explore all tools that specialize in translate code. This domain focus ensures StarCoder2-15B delivers optimized results for this specific requirement.
Explore all tools that specialize in debug code. This domain focus ensures StarCoder2-15B delivers optimized results for this specific requirement.
Explore all tools that specialize in refactor code. This domain focus ensures StarCoder2-15B delivers optimized results for this specific requirement.
Explore all tools that specialize in code completion. This domain focus ensures StarCoder2-15B delivers optimized results for this specific requirement.
Improves inference speed and reduces memory consumption by grouping query vectors.
Allows the model to attend to a larger context window (16,384 tokens) with reduced computational cost by focusing on a sliding window of 4,096 tokens.
The model is trained to fill in missing code segments, enhancing its ability to understand and generate code in various contexts.
Supports 8-bit and 4-bit quantization using bitsandbytes, reducing memory footprint and enabling deployment on resource-constrained devices.
Seamlessly scales training and inference across multiple GPUs using the Accelerate library.
Install transformers from source: pip install git+https://github.com/huggingface/transformers.git
Install necessary libraries: pip install accelerate bitsandbytes
Load the tokenizer: tokenizer = AutoTokenizer.from_pretrained("bigcode/starcoder2-15b")
Load the model: model = AutoModelForCausalLM.from_pretrained("bigcode/starcoder2-15b", device_map="auto", torch_dtype=torch.bfloat16)
Encode input: inputs = tokenizer.encode("def print_hello_world():", return_tensors="pt").to("cuda")
Generate output: outputs = model.generate(inputs)
Decode output: print(tokenizer.decode(outputs[0]))
All Set
Ready to go
Verified feedback from other users.
"Users praise its code generation capabilities, but some note its limitations in complex tasks."
Post questions, share tips, and help other users.

Google's family of multimodal AI models.

A minimal code editor crafted for speed and collaboration with humans and AI.
Unlock the power of AI models for various applications with a scalable and flexible API.

Ultra-fast, low-memory AI productivity and refactoring engine for Visual Studio.

The AI-driven code assistant for streamlined development and instant technical documentation.

The first AI code assistant for enterprise-grade development with total privacy control.