TRACCERR/gemma-4-E4B-it-FT-GGUF

Uploaded finetuned model

• Developed by: Prompt48
• License: apache-2.0
• Finetuned from model : unsloth/gemma-4-E4B-it

This gemma4 model was trained 2x faster with Unsloth and Huggingface's TRL library.

Training Dataset

This model was fine-tuned on the Bitext Customer Support LLM Chatbot Training Dataset.

• Creator: Bitext Innovations
• Dataset License: CDLA-Sharing-1.0
• Language: English (en-US)
• Size: 26,872 question/answer pairs (~1,000 per intent), 3.57 million total tokens
• Format: Each sample contains: flags, instruction, category, intent, response

The dataset covers 10 categories and 27 intents for customer support chatbot applications: ACCOUNT (create, delete, edit, switch), CANCELLATION_FEE, DELIVERY (options), FEEDBACK (complaint, review), INVOICE (check, get), NEWSLETTER (subscription), ORDER (cancel, change, place), PAYMENT (check methods, payment issue), REFUND (check policy, track), SHIPPING_ADDRESS (change, set up).

It includes 12 linguistic variation tags capturing lexical, syntactic, register, and stylistic diversity, and 30 entity/slot placeholder types. Applicable across 20+ industry verticals (retail, banking, healthcare, travel, etc.).

GGUF Quantizations

Filename	Quantization	Size	BPW	Description
gemma-4-e4b-it-ft-IQ4_NL.gguf	IQ4_NL	4.9 GB	5.53	Good quality, recommended for most use cases
gemma-4-e4b-it-ft-IQ3_M.gguf	IQ3_M	4.4 GB	5.00	Smaller size, slightly reduced quality
gemma-4-e4b-it-ft-Q4_K_M.gguf	Q4_K_M	5.0 GB	5.66	K-quant medium, good balance of quality and size

Original Safetensors

The original finetuned safetensors model files are also included in this repository.

Source

Quantized from Prompt48/gemma-4-E4B-it-FT