ai-colombia/gemma4-e4b-job-searcher-gguf

Gemma 4 E4B — AI Job Searcher (GGUF Q4_K_M)

Fine-tuned google/gemma-4-E4B-it for multilingual job search assistance. Quantized to Q4_K_M for efficient local inference via llama.cpp.

Model Details

Property	Value
Base model	google/gemma-4-E4B-it
Fine-tune method	LoRA (r=16, α=16)
Training hardware	NVIDIA RTX 5080 (16GB VRAM)
Quantization	Q4_K_M (~5 GB)
Format	GGUF (llama.cpp compatible)
Languages	EN, ES, FR, DE, PT, NO, DA, FI, SV
Task	Job search assistance, CV help, interview prep

Training

• Dataset: ai-colombia/ai-job-searcher-finetune-data
• Training steps: 148
• Effective batch size: 16 (batch 2 × grad_accum 8)
• Epochs: 2
• Learning rate: 5e-5
• Max sequence length: 2048
• Precision: bf16

LoRA Adapter

The unmerged LoRA adapter is available at: ai-colombia/gemma4-e4b-job-searcher-lora

Usage

llama.cpp / LM Studio / Ollama

./llama-cli -m gemma4-e4b-job-searcher-q4_k_m.gguf --chat-template gemma -p "You are a helpful job search assistant." -i

Python (llama-cpp-python)

from llama_cpp import Llama

llm = Llama(
    model_path="gemma4-e4b-job-searcher-q4_k_m.gguf",
    n_ctx=2048,
    n_gpu_layers=-1,  # use GPU if available
)

response = llm.create_chat_completion(
    messages=[
        {"role": "system", "content": "You are a helpful job search assistant."},
        {"role": "user", "content": "Help me write a cover letter for a software engineer position."}
    ]
)
print(response["choices"][0]["message"]["content"])

Capabilities

• Job search guidance — advice on finding jobs, job boards, networking
• CV / Resume writing — structure, content, ATS optimization tips
• Cover letter writing — tailored letters for specific roles
• Interview preparation — common questions, STAR method, salary negotiation
• Career advice — career transitions, skill gaps, industry insights
• Multilingual — responds in EN, ES, FR, DE, PT, NO, DA, FI, SV

Limitations

• Based on google/gemma-4-E4B-it — subject to Gemma's usage policy
• Knowledge cutoff from base model training data
• Q4_K_M quantization may reduce quality on complex reasoning vs the full fp16 model
• Not suitable for real-time job listings (no web access)

License

This model is subject to the Gemma Terms of Use. The fine-tuning data and LoRA weights are released under Apache 2.0.