---

base_model: Qwen/Qwen3.5-4B library_name: transformers license: apache-2.0 language: - en pipeline_tag: text-generation tags: - qwen3.5 - coding - reasoning - lora - sft - agentic - python - javascript - sql base_model_relation: finetune

aiAI_coder_V1.4B

4B parameters • Fine-tuned for coding & agentic tasks • <1 hour training

🤗 Model | 📊 Evaluation

Model Overview

aiAI_coder_V1.4B is a specialized coding and agentic assistant fine-tuned from Qwen/Qwen3.5-4B. It is designed to excel in:

• Multi-language Code Generation: Python, JavaScript, TypeScript, and SQL
• Reasoning & Problem-Solving: Step-by-step thinking with <think> tag support
• Agentic Workflows: Tool calling, multi-turn interactions, and task completion
• Instruction Adherence: Following complex, constrained prompts with high accuracy
• Cost Efficiency: Optimized for low-latency inference on consumer hardware

This model was distilled from high-quality Grok 4.6 completions and trained with a highly efficient Supervised Fine-Tuning (SFT) recipe, achieving strong coding benchmark performance at a fraction of the cost of larger models [citation:1][citation:7].

Model Details

Model Description

aiAI_coder_V1.4B is an instruction-tuned language model optimized for code synthesis, debugging, and agentic assistance. It supports:

• Fast, deterministic responses for coding tasks
• Accurate code generation in Python, JavaScript, TypeScript, and SQL
• Multi-turn reasoning with explicit thinking separation (<think>...</think>)
• Native support for tool calling and structured outputs

The model can be used as a lightweight, cost-effective alternative to frontier models in many developer workflows.

• Developed by: [aiAI]
• Funded by: [nitrous-0xide (owner & founder)]
• Model type: Text-generation / Instruction-following
• Language(s): English
• License: Apache-2.0
• Finetuned from: Qwen/Qwen3.5-4B [citation:1][citation:10]

Uses

Direct Use

The model can be used as-is for:

• Interactive coding assistants and chatbots
• Code completion and debugging in IDEs
• Generating unit tests and documentation
• SQL query generation and optimization
• Agentic workflows requiring planning and tool use [citation:1]

Out-of-Scope Use

• Generating malicious code or content that violates applicable laws
• Real-time decision-making in safety-critical systems
• Any use that violates the Apache-2.0 license

Bias, Risks, and Limitations

• Hallucination: May occasionally produce plausible but incorrect code or explanations
• Security: Generated code should be reviewed for security vulnerabilities
• Context Window: While optimized for 262K context, performance may degrade at extreme lengths [citation:7]
• Language Coverage: Primarily trained on English data; performance on other languages is limited

Recommendations

• Human-in-the-loop review of generated code before deployment
• Use explicit safety filters for disallowed content
• Test outputs in sandboxed environments when executing generated code

How to Get Started with the Model

Load with Transformers (BF16 checkpoint)

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "aiAI_coder_V1.4B"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

messages = [
    {"role": "system", "content": "You are a helpful coding assistant. Think step by step."},
    {"role": "user", "content": "Write a Python function to reverse a linked list in-place."}
]

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

outputs = model.generate(
    inputs,
    max_new_tokens=1024,
    temperature=0.2,
    do_sample=True
)

response = tokenizer.decode(outputs[0][inputs.shape[1]:], skip_special_tokens=True)
print(response)