Modelpublic

ardakalayci/bist-tool-gemma4-e2b

sourceHugging Facegemmaupdated 2mo agoView on Hugging Face

0likes15downloads

Model Card

BIST Tool — Gemma 4 E2B

Fine-tuned Gemma 4 E2B for financial tool-calling (BIST / US / FX / crypto / funds-style tools).

—Base: `unsloth/gemma-4-E2B-it`
—Training: Unsloth Studio, Gemma-4 chat template, thinking disabled
—Data: `bist-tool-sft` (update username after upload)

Behavior

Intent	Tools
Known ticker price (`GARAN`, `THYAO`, `AAPL`)	`get_quote`
Research a known stock	`get_quote` + `get_profile` + `get_technical_analysis`
Unknown name (`EUREN`)	4× `search_symbol` (bist, us, fund, crypto) — not EUR/FX
Explicit FX (`euro kuru`, `dolar kaç`)	`get_quote` with `market: fx`

Example (Gemma 4 tool tokens):

text

<|tool_call>call:get_quote{symbol:<|"|>THYAO<|"|>,market:<|"|>bist<|"|>}<tool_call|>

Usage

python

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

REPO = "HF_USER/bist-tool-gemma4-e2b"

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

Use the Gemma 4 chat template. Prefer thinking off for clean tool tokens. If the host prefixes tool names (mcp__…__get_quote), map them to short labels (get_quote, search_symbol, …).

License

Derived from Gemma weights; subject to the Gemma license and terms.