40 lines
988 B
Markdown
40 lines
988 B
Markdown
|
---
|
||
|
|
license: other
|
||
|
|
base_model: Qwen/Qwen3-8B
|
||
|
|
library_name: transformers
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
tags:
|
||
|
|
- qwen3
|
||
|
|
- mapfinben
|
||
|
|
- finance
|
||
|
|
- lora
|
||
|
|
- merged
|
||
|
|
---
|
||
|
|
|
||
|
|
# Qwen3-8B_MapFinBen
|
||
|
|
|
||
|
|
Merged Qwen3-8B model fine-tuned for the MapFinBen financial benchmark.
|
||
|
|
|
||
|
|
This repository contains the full merged Hugging Face model weights and tokenizer files. It also includes Ollama-compatible template files for the benchmark conversion workflow.
|
||
|
|
|
||
|
|
## Evaluation Template
|
||
|
|
|
||
|
|
For the official GGUF/Ollama workflow, use `template.tmpl`. It defines a simple ChatML prompt format and avoids Qwen3 thinking/tool templates:
|
||
|
|
|
||
|
|
```text
|
||
|
|
<|im_start|>user
|
||
|
|
{{ prompt }}
|
||
|
|
<|im_end|>
|
||
|
|
<|im_start|>assistant
|
||
|
|
```
|
||
|
|
|
||
|
|
Recommended deterministic evaluation settings:
|
||
|
|
|
||
|
|
```text
|
||
|
|
temperature 0.0
|
||
|
|
top_p 1.0
|
||
|
|
seed 42
|
||
|
|
stop: <|im_end|>, <|endoftext|>
|
||
|
|
```
|
||
|
|
|
||
|
|
`Modelfile` is provided as a manual Ollama import reference. The benchmark organizer may regenerate the base Modelfile after HF-to-GGUF conversion and append `template.tmpl`.
|