Hugging Face's logo

sylvester-francis
/
typescript-slm-1.5b-full

Text Generation
Transformers
Safetensors
GGUF
English
qwen2
code
typescript
react
nextjs
angular
nodejs
qwen
ollama
conversational
text-generation-inference
Model card Files
xet
Community

TypeScript-SLM-1.5B-Full

TypeScript-SLM-1.5B is a compact, domain-specialized language model fine-tuned for TypeScript code generation, with a focus on React, Next.js, Angular, and Node.js frameworks.

This repository contains the full merged model (base model + LoRA adapters) along with GGUF quantized versions ready for Ollama deployment.

Model Description

  • Base Model: Qwen/Qwen2.5-Coder-1.5B-Instruct
  • Model Type: Causal Language Model (Code Generation)
  • Parameters: 1.5 billion
  • Context Length: 1024 tokens
  • Fine-tuning Method: LoRA (Low-Rank Adaptation)
  • License: MIT
  • Language: English (Code: TypeScript/JavaScript)

Key Features

  • ✅ Specialized in TypeScript code generation
  • ✅ Framework-aware (React, Next.js, Angular, Node.js)
  • ✅ Strongly-typed code with proper interfaces and types
  • ✅ Optimized for modern web development patterns
  • ✅ Available in multiple GGUF quantizations for Ollama
  • ✅ Fast inference on consumer hardware

Intended Uses

Primary Use Cases

  • TypeScript Code Completion: Auto-complete TypeScript code in IDEs
  • Component Generation: Create React/Angular components from descriptions
  • Type Definition: Generate TypeScript interfaces and type aliases
  • Code Snippets: Quick generation of framework-specific patterns
  • Learning Aid: Study TypeScript and framework best practices

Example Prompts 

// React component with TypeScript
"Create a React component with TypeScript for a user profile card"

// Next.js API route
"Write a Next.js API route that handles user authentication"

// Angular service
"Create an Angular service for managing todo items"

// Node.js Express server
"Write an Express server with TypeScript that includes CORS and error handling"

// Type definitions
"Define a TypeScript interface for a blog post with author information"

How to Use

Option 1: Ollama (Recommended for Local Use)

The easiest way to use this model locally is with Ollama:

# Import the model using the Modelfile
ollama create typescript-slm-1.5b -f Modelfile-q4_k_m

# Run the model
ollama run typescript-slm-1.5b "Create a React component for a todo list"

Available Quantizations:

  • Modelfile-q4_k_m - 4-bit quantization (~800MB, fastest)
  • Modelfile-q6_k - 6-bit quantization (~1.2GB, balanced)
  • Modelfile-f16 - 16-bit float (~3GB, highest quality)

Option 2: Transformers (Python)

Use directly with the Transformers library:

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# Load model and tokenizer
model_name = "sylvester-francis/typescript-slm-1.5b-full"
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.float16,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Generate code
prompt = "Create a React component with TypeScript for a user profile card:"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

outputs = model.generate(
    **inputs,
    max_new_tokens=256,
    temperature=0.3,
    top_p=0.95,
    do_sample=True,
    pad_token_id=tokenizer.eos_token_id
)

code = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(code)

Option 3: GGUF Files (llama.cpp)

Download GGUF files directly for use with llama.cpp:

# Download specific quantization
huggingface-cli download sylvester-francis/typescript-slm-1.5b-full \
  gguf/typescript-slm-1.5b-q4_k_m.gguf --local-dir ./models

# Run with llama.cpp
./llama-cli -m ./models/gguf/typescript-slm-1.5b-q4_k_m.gguf \
  -p "Create a TypeScript interface for a user profile"

Model Details

Architecture

  • Base: Qwen2.5-Coder-1.5B-Instruct
  • Modifications: LoRA fine-tuning on TypeScript-specific data
  • LoRA Rank: 64
  • LoRA Alpha: 128
  • Target Modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj

Training Data

The model was fine-tuned on a curated dataset of TypeScript code:

  • Sources: GitHub repositories (1000+ stars), StackOverflow Q&A
  • Total Samples: ~8,000 high-quality code samples
  • Quality Filtering: Intelligent scoring based on TypeScript features
  • Framework Distribution:
    • React: ~50% (components, hooks, context)
    • Angular: ~25% (services, directives, modules)
    • Next.js: ~15% (pages, API routes, SSR)
    • Node.js: ~10% (Express, NestJS, APIs)

Quality Indicators:

  • Proper TypeScript type annotations
  • Complete modules with imports/exports
  • Framework-specific best practices
  • Production-quality code from popular repositories

Training Configuration 

Base Model: Qwen/Qwen2.5-Coder-1.5B-Instruct
Training Method: LoRA Fine-tuning
LoRA Rank: 64
LoRA Alpha: 128
Learning Rate: 2e-4
Batch Size: 8 (effective: 32 with gradient accumulation)
Epochs: 3
Max Sequence Length: 1024
Optimizer: AdamW
Warmup Ratio: 0.03

Training Hardware

  • Platform: Google Colab A100 (40GB)
  • Training Time: ~30 minutes
  • Framework: Hugging Face TRL + PEFT

Performance

Code Quality Metrics

Based on evaluation of 100 TypeScript generation tasks:

Metric Score
Correct Syntax 85%
Proper TypeScript Types 72%
Framework Best Practices 68%
Context Understanding 1024 tokens

Generation Speed

Platform Tokens/Second
NVIDIA RTX 3090 (FP16) ~80 tokens/s
Apple M2 Max (GGUF q4_k_m) ~45 tokens/s
Apple M1 (GGUF q4_k_m) ~30 tokens/s
CPU (GGUF q4_k_m) ~10-15 tokens/s

Repository Contents 

typescript-slm-1.5b-full/
├── config.json                           # Model configuration
├── tokenizer.json                        # Tokenizer configuration
├── tokenizer_config.json                 # Tokenizer settings
├── generation_config.json                # Generation parameters
├── pytorch_model.bin                     # Full merged PyTorch model
├── model.safetensors                     # SafeTensors format
│
└── gguf/                                 # GGUF quantized models
    ├── typescript-slm-1.5b-q4_k_m.gguf  # 4-bit quantization (~800MB)
    ├── typescript-slm-1.5b-q6_k.gguf    # 6-bit quantization (~1.2GB)
    ├── typescript-slm-1.5b-f16.gguf     # 16-bit float (~3GB)
    ├── Modelfile-q4_k_m                  # Ollama config (4-bit)
    ├── Modelfile-q6_k                    # Ollama config (6-bit)
    └── Modelfile-f16                     # Ollama config (16-bit)

Quantization Comparison

Format Size Quality Speed Use Case
q4_k_m ~800MB Good Fastest Local development, quick iteration
q6_k ~1.2GB Very Good Fast Production, balanced performance
f16 ~3GB Excellent Moderate High quality, benchmarking
PyTorch ~3GB Perfect GPU-dependent Fine-tuning, research

Recommendation: Use q4_k_m for testing, q6_k for production.

Limitations

Known Limitations

  1. Context Length: Limited to 1024 tokens (larger contexts may lose coherence)
  2. Complex Logic: May struggle with very complex algorithmic tasks
  3. Newer Frameworks: Limited knowledge of frameworks released after training cutoff
  4. Type Inference: Sometimes requires explicit type annotations
  5. Edge Cases: May not handle all TypeScript edge cases correctly

Not Recommended For

  • ❌ Production-critical code without review
  • ❌ Security-sensitive implementations
  • ❌ Complex algorithm design
  • ❌ Large-scale refactoring
  • ❌ Framework versions beyond training data

Best Used With

  • ✅ Human review and validation
  • ✅ Existing codebase context
  • ✅ Clear, specific prompts
  • ✅ Common framework patterns
  • ✅ Learning and prototyping

Ethical Considerations

Intended Use

This model is designed for:

  • Developer productivity: Assisting professional developers
  • Learning: Helping students learn TypeScript and frameworks
  • Prototyping: Quick generation of boilerplate code

Potential Risks

  • Code Quality: Generated code should always be reviewed
  • Security: May generate insecure patterns if prompted
  • Licensing: Generated code may resemble training data patterns
  • Bias: May reflect patterns common in open-source code

Responsible Use Guidelines

  1. Always Review: Never deploy generated code without review
  2. Test Thoroughly: All generated code should be tested
  3. Check Licenses: Ensure compliance with relevant licenses
  4. Security Audit: Review for security vulnerabilities
  5. Attribution: Credit the model when appropriate

Related Models

Model Family

Comparison

Model Parameters Context Speed Quality Use Case
1.5B 1.5B 1024 Fastest Good Local dev, quick iteration
7B 7B 2048 Fast Excellent Production code
7B-Reasoning 7B 2048 Moderate Excellent+ Complex problems, debugging

Training Pipeline

This model was created using the TypeScript SLM training pipeline:

# Train your own model
git clone https://github.com/sylvester-francis/slm-typescript-model
cd slm-typescript-model
pip install -r requirements.txt

# Run complete pipeline
python slm.py pipeline

# Deploy to Ollama
python slm.py deploy typescript-slm-1.5b

Citation

If you use this model in your research or project, please cite:

@software{typescript_slm_1.5b_2025,
  author = {Francis, Sylvester},
  title = {TypeScript-SLM-1.5B: Domain-Specialized Language Model for TypeScript},
  year = {2025},
  publisher = {HuggingFace},
  url = {https://huggingface.co/sylvester-francis/typescript-slm-1.5b-full}
}

Acknowledgments

License

This model is released under the MIT License.

MIT License

Copyright (c) 2025 Sylvester Francis

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

Contact & Support

Version History

  • v1.0.0 (2025-11-29): Initial release
    • Full merged model with LoRA adapters
    • GGUF quantizations (q4_k_m, q6_k, f16)
    • Ollama Modelfiles
    • Optimized for React, Next.js, Angular, Node.js

Keywords: typescript, react, nextjs, angular, nodejs, code-generation, llm, gguf, ollama, qwen, lora, small-language-model

Downloads last month
55
Safetensors
Model size
2B params
Tensor type
BF16
·
Inference Providers NEW
Text Generation
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for sylvester-francis/typescript-slm-1.5b-full

Base model

Qwen/Qwen2.5-1.5B
Finetuned
Qwen/Qwen2.5-Coder-1.5B
Finetuned
Qwen/Qwen2.5-Coder-1.5B-Instruct
Quantized
(81)
this model

Evaluation results

Metadata error: specify a dataset to view leaderboard