GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
RCaz
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
RCaz
Model Tree
Qwen/Qwen2.5-7B-Instruct
Input Modalities
Output Modalities
Supported Functionality
The model has expert knowledge across all 6 EU framework programmes:
| Programme | Period | Projects |
|---|---|---|
| FP4 | 1994–1998 | ~13K |
| FP5 | 1998–2002 | ~17K |
| FP6 | 2002–2006 | ~10K |
| FP7 | 2007–2013 | ~26K |
| Horizon 2020 | 2014–2020 | ~36K |
| Horizon Europe | 2021–2027 | ~24K |
It can answer questions about:
python
from peft import PeftModelfrom transformers import AutoModelForCausalLM, AutoTokenizer base_model = AutoModelForCausalLM.from_pretrained( "Qwen/Qwen2.5-7B-Instruct", torch_dtype="auto", device_map="auto",)model = PeftModel.from_pretrained(base_model, "RCaz/Qwen2.5-7B-EU-Funding-Expert")tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-7B-Instruct") messages = [ {"role": "system", "content": "You are an expert assistant specializing in European Union research funding programmes."}, {"role": "user", "content": "What were the main funding priorities under Horizon 2020?"}] text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)inputs = tokenizer(text, return_tensors="pt").to(model.device)outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.7)print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))| Parameter | Value |
|---|---|
| Base model | Qwen/Qwen2.5-7B-Instruct (7.62B params) |
| Method | LoRA SFT (Supervised Fine-Tuning) |
| LoRA rank | 64 |
| LoRA alpha | 16 |
| LoRA target | all-linear layers |
| LoRA dropout | 0.05 |
| Trainable params | ~410M (5.4% of total) |
| Dataset | RCaz/eu-funding-cordis-qa |
The training data was generated from official CORDIS CSV exports covering 126K+ EU-funded research projects. Eight types of Q&A conversations were created:
All conversations follow the ChatML format with a system prompt establishing EU funding expertise.
If you use this model, please cite:
bibtex
@misc{rcaz2026eufunding, title={Qwen2.5-7B-EU-Funding-Expert: A Fine-tuned LLM for European Research Funding}, author={RCaz}, year={2026}, url={https://huggingface.co/RCaz/Qwen2.5-7B-EU-Funding-Expert}}| Training samples | 413,499 |
| Validation samples | 21,764 |
| Max sequence length | 2048 |
| Packing | BFD (Best-Fit Decreasing) |
| Avg tokens/sample | ~285 |
| Precision | bf16 |
| Optimizer | AdamW (fused) |
| Learning rate | 2e-4 (cosine schedule) |
| Batch size | 1 × 8 grad accum = 8 effective |
| Hardware | NVIDIA L4 (24GB) |
| Flash Attention | 2.0 |
| Gradient checkpointing | ✓ |
from peft import PeftModelfrom transformers import AutoModelForCausalLM, AutoTokenizer base_model = AutoModelForCausalLM.from_pretrained( "Qwen/Qwen2.5-7B-Instruct", torch_dtype="auto", device_map="auto",)model = PeftModel.from_pretrained(base_model, "RCaz/Qwen2.5-7B-EU-Funding-Expert")tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-7B-Instruct") messages = [ {"role": "system", "content": "You are an expert assistant specializing in European Union research funding programmes."}, {"role": "user", "content": "What were the main funding priorities under Horizon 2020?"}] text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)inputs = tokenizer(text, return_tensors="pt").to(model.device)outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.7)print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))@misc{rcaz2026eufunding, title={Qwen2.5-7B-EU-Funding-Expert: A Fine-tuned LLM for European Research Funding}, author={RCaz}, year={2026}, url={https://huggingface.co/RCaz/Qwen2.5-7B-EU-Funding-Expert}}