---
language:
- en
license: apache-2.0
base_model: Qwen/Qwen2.5-7B-Instruct
pipeline_tag: text-generation
library_name: transformers
tags:
- fundraising
- non-profit
- coaching
- conversational
- sft
- lora
datasets:
- Mic-Fundraiser/fundraising-english-sft
---

# Fundraising Coach — English

A conversational language model fine-tuned exclusively on nonprofit fundraising
coaching in English. Produces structured answers: **diagnosis → options → recommendation**.
Trained to avoid fabricating statistics or specific legal/tax guidance.

Base: [`Qwen/Qwen2.5-7B-Instruct`](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct).
Dataset: [`Mic-Fundraiser/fundraising-english-sft`](https://huggingface.co/datasets/Mic-Fundraiser/fundraising-english-sft) (10,085 train + 530 val, distilled from 14 reference texts).

## Try it

[Live demo on Hugging Face Spaces](https://huggingface.co/spaces/Mic-Fundraiser/fundraising-coach-en-demo)

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "Mic-Fundraiser/fundraising-coach-en"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype="auto", device_map="auto")

SYSTEM = (
    "You are an experienced nonprofit fundraising coach. For any substantive question, "
    "structure your answer as: (1) a brief diagnosis of what the core issue really is, "
    "(2) the main options with their tradeoffs, (3) a concrete recommendation with the "
    "reasoning behind it. Be direct, grounded, and practical."
)
messages = [
    {"role": "system", "content": SYSTEM},
    {"role": "user", "content": "Our annual fund is flat. What should I look at first?"},
]
input_ids = tokenizer.apply_chat_template(messages, add_generation_prompt=True,
                                          return_tensors="pt", return_dict=False).to(model.device)
out = model.generate(input_ids, max_new_tokens=500, do_sample=True,
                     temperature=0.6, top_p=0.9, repetition_penalty=1.05)
print(tokenizer.decode(out[0][input_ids.shape[1]:], skip_special_tokens=True))
```

## Training

| | |
|---|---|
| Method | SFT + QLoRA (4-bit) |
| LoRA | r=32, α=64, dropout=0.05 |
| Modules | q/k/v/o/gate/up/down |
| Epochs | 2 |
| LR | 1.5e-4 · cosine |
| Eff. batch | 16 · grad accum 8 |
| Seq length | 2048 · packing enabled |
| Hardware | NVIDIA L40S (HF Jobs) |
| Framework | TRL + PEFT + transformers |

## Limitations

- Knowledge is shaped by the 14 source texts; regulation, platforms, and trends post-2023 are underrepresented.
- Deliberately refuses to invent precise statistics. May feel evasive on "what percentage of donors..." questions.
- English-only. For Italian see [`fundraising-assistant`](https://huggingface.co/Mic-Fundraiser/fundraising-assistant).
- 2048-token context during training: long multi-turn sessions may lose earlier details.