48 lines
1.6 KiB
Markdown
48 lines
1.6 KiB
Markdown
|
---
|
||
|
|
base_model: unsloth/gemma-3-270m-it
|
||
|
|
tags:
|
||
|
|
- text-generation-inference
|
||
|
|
- transformers
|
||
|
|
- unsloth
|
||
|
|
- gemma3_text
|
||
|
|
license: apache-2.0
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
---
|
||
|
|
|
||
|
|
---
|
||
|
|
license: apache-2.0
|
||
|
|
base_model: google/gemma-3-270m-it
|
||
|
|
library_name: transformers
|
||
|
|
tags:
|
||
|
|
- unsloth
|
||
|
|
- gemma-3
|
||
|
|
- emoji
|
||
|
|
- translation
|
||
|
|
- multilingual
|
||
|
|
- gguf
|
||
|
|
- edge-ai
|
||
|
|
datasets:
|
||
|
|
- custom-curated-emoji-distillation
|
||
|
|
---
|
||
|
|
|
||
|
|
# 🚀 Emojify-300M (Gemma-3-270M Fine-tuned)
|
||
|
|
|
||
|
|
**Emojify-300M** is a hyper-compact, specialized Large Language Model (LLM) designed for **semantic text-to-emoji distillation**. Based on the **Gemma-3-270M-IT** architecture, this model is optimized for edge computing and ultra-low latency applications.
|
||
|
|
|
||
|
|
## 🛠 Technical Specifications
|
||
|
|
- **Architecture:** Gemma-3 (270M parameters)
|
||
|
|
- **Fine-tuning Method:** LoRA (Low-Rank Adaptation) via **Unsloth**
|
||
|
|
- **Context Window:** 2048 tokens
|
||
|
|
- **Language Support:** Native Multilingual
|
||
|
|
|
||
|
|
## 📈 Performance & Benchmarks (Local CPU)
|
||
|
|
Tested on consumer-grade hardware (e.g., laptop CPU):
|
||
|
|
- **Prompt Evaluation:** ~210 tokens/s
|
||
|
|
- **Token Generation (Eval):** ~48 tokens/s
|
||
|
|
- **Total Latency:** < 500ms (near-instant response)
|
||
|
|
|
||
|
|
## 🎯 Key Features
|
||
|
|
- **Zero-Shot Cross-Lingual Transfer:** Leveraging Gemma 3's robust base weights, the model accurately processes languages not explicitly present in the fine-tuning set.
|
||
|
|
- **Noise Suppression:** Specifically trained to inhibit conversational filler and "hallucinated" text, focusing strictly on relevant emoji output.
|
||
|
|
- **EOS Stability:** Fine-tuned to respect End-of-Sequence (EOS) tokens, preventing the common "looping" behavior seen in smaller models.
|