ManideepVarma003/finexpert-qwen3-4b-gguf overview
FinExpert, GGUF Q4 K M The FinExpert LoRA adapter https://huggingface.co/ManideepVarma003/finexpert qwen3 4b lora merged into Qwen3 4B Instruct 2507 and quanti…
Runs locally from ~2.33 GB disk (4 GB VRAM class GPUs with llama.cpp / guIDE).
Repository Files & Downloads
| File | Type | Quantization | Size | Link |
|---|---|---|---|---|
| finexpert-qwen3-4b-Q4_K_M.gguf | GGUF | Q4_K_M | 2.33 GB | Download |
Model Details
| Model ID | ManideepVarma003/finexpert-qwen3-4b-gguf |
|---|---|
| Author | ManideepVarma003 |
| Pipeline | — |
| License | apache-2.0 |
| Base model | unsloth/Qwen3-4B-Instruct-2507-unsloth-bnb-4bit |
| Last modified | 2026-10-15T11:24:48.000Z |
Model README
---
base_model: unsloth/Qwen3-4B-Instruct-2507-unsloth-bnb-4bit
license: apache-2.0
tags: [gguf, llama.cpp, finance, qlora, finqa, tat-qa]
---
FinExpert, GGUF (Q4_K_M)
The FinExpert LoRA adapter merged into Qwen3-4B-Instruct-2507 and quantised to
Q4_K_M for CPU inference with llama.cpp. Evaluation, intended use and limitations: see the adapter's model card;
code and experiment log: <https://github.com/mani1234765/finexpert>.
Conversion check. On 40 fixed validation questions (20 FinQA + 20 TAT-QA), the GGUF gave the same final answer as the evaluated adapter on 25/40; correct answers: GGUF 22/40 vs adapter 24/40. CPU speed on Kaggle (4 threads): about 45.8 s per answer.
Reported benchmark results belong to the adapter on the 4-bit base; this file is a convenience conversion for the demo.
Prompt format: ChatML with the FinExpert system prompt; the exact builder is in the adapter's quick start.
Use greedy decoding (temperature 0) and stop at <|im_end|>.
Run ManideepVarma003/finexpert-qwen3-4b-gguf with guIDE
Download guIDE — the AI-native code editor with local LLM inference and 69 built-in tools.
Source: Hugging Face · Compare models