Instructions to use grc-iit/FunctionGemma-ndp with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use grc-iit/FunctionGemma-ndp with Transformers:
# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("grc-iit/FunctionGemma-ndp", dtype="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps
- Unsloth Studio
How to use grc-iit/FunctionGemma-ndp with Unsloth Studio:
Install Unsloth Studio (macOS, Linux, WSL)
curl -fsSL https://unsloth.ai/install.sh | sh # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for grc-iit/FunctionGemma-ndp to start chatting
Install Unsloth Studio (Windows)
irm https://unsloth.ai/install.ps1 | iex # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for grc-iit/FunctionGemma-ndp to start chatting
Using HuggingFace Spaces for Unsloth
# No setup required # Open https://huggingface.co/spaces/unsloth/studio in your browser # Search for grc-iit/FunctionGemma-ndp to start chatting
Load model with FastModel
pip install unsloth from unsloth import FastModel model, tokenizer = FastModel.from_pretrained( model_name="grc-iit/FunctionGemma-ndp", max_seq_length=2048, )
Add live-MCP test code block
Browse files
README.md
CHANGED
|
@@ -49,6 +49,80 @@ Output format is FunctionGemma native:
|
|
| 49 |
<start_function_call>call:list_organizations{server:<escape>global<escape>}<end_function_call>
|
| 50 |
```
|
| 51 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
## Files
|
| 53 |
|
| 54 |
- `merged_16bit/` — full safetensors checkpoint
|
|
|
|
| 49 |
<start_function_call>call:list_organizations{server:<escape>global<escape>}<end_function_call>
|
| 50 |
```
|
| 51 |
|
| 52 |
+
## Live test against the upstream NDP MCP
|
| 53 |
+
|
| 54 |
+
End-to-end: model → tool call → upstream `clio-kit` NDP MCP → real NDP response.
|
| 55 |
+
|
| 56 |
+
```python
|
| 57 |
+
# /// script
|
| 58 |
+
# requires-python = ">=3.11"
|
| 59 |
+
# dependencies = [
|
| 60 |
+
# "transformers>=4.45", "torch>=2.4", "accelerate>=0.34",
|
| 61 |
+
# "sentencepiece>=0.2", "protobuf>=4", "mcp>=1.0",
|
| 62 |
+
# ]
|
| 63 |
+
# ///
|
| 64 |
+
import asyncio, json, re
|
| 65 |
+
import torch
|
| 66 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 67 |
+
from mcp import ClientSession, StdioServerParameters
|
| 68 |
+
from mcp.client.stdio import stdio_client
|
| 69 |
+
|
| 70 |
+
MID = "shazzadulimun/FunctionGemma-ndp"
|
| 71 |
+
PROMPT = "List all organizations on the NDP global server"
|
| 72 |
+
|
| 73 |
+
# 14-tool NDP catalog reshaped as OpenAI function specs (truncated here).
|
| 74 |
+
tools = [{"type": "function", "function": {
|
| 75 |
+
"name": "list_organizations",
|
| 76 |
+
"description": "List organizations available in the National Data Platform.",
|
| 77 |
+
"parameters": {"type": "object", "properties": {
|
| 78 |
+
"name_filter": {"type": "string"}, "server": {"type": "string"},
|
| 79 |
+
}, "required": []},
|
| 80 |
+
}}]
|
| 81 |
+
|
| 82 |
+
tok = AutoTokenizer.from_pretrained(MID, subfolder="merged_16bit")
|
| 83 |
+
mdl = AutoModelForCausalLM.from_pretrained(
|
| 84 |
+
MID, subfolder="merged_16bit", dtype=torch.bfloat16, device_map="auto",
|
| 85 |
+
)
|
| 86 |
+
text = tok.apply_chat_template(
|
| 87 |
+
[{"role": "user", "content": PROMPT}],
|
| 88 |
+
tools=tools, add_generation_prompt=True, tokenize=False,
|
| 89 |
+
)
|
| 90 |
+
inp = tok(text, return_tensors="pt").to(mdl.device)
|
| 91 |
+
out = mdl.generate(**inp, max_new_tokens=300)
|
| 92 |
+
raw = tok.decode(out[0][inp.input_ids.shape[-1]:], skip_special_tokens=False)
|
| 93 |
+
|
| 94 |
+
# Parse FunctionGemma format: <start_function_call>call:NAME{k:v,...}<end_function_call>
|
| 95 |
+
m = re.search(r"<start_function_call>\s*call:(\w+)\s*\{(.*?)\}\s*<end_function_call>",
|
| 96 |
+
raw, re.DOTALL)
|
| 97 |
+
name = m.group(1)
|
| 98 |
+
args = {}
|
| 99 |
+
for k, v in re.findall(r"(\w+)\s*:\s*(<escape>.*?<escape>|None|\w+)", m.group(2)):
|
| 100 |
+
if v == "None":
|
| 101 |
+
continue # strip phantom nulls
|
| 102 |
+
args[k] = re.sub(r"<escape>|<escape>", "", v) if "<escape>" in v else v
|
| 103 |
+
|
| 104 |
+
# Spawn the upstream clio-kit NDP MCP and call the parsed tool against it.
|
| 105 |
+
async def call():
|
| 106 |
+
params = StdioServerParameters(command="uvx", args=[
|
| 107 |
+
"--from",
|
| 108 |
+
"git+https://github.com/iowarp/clio-kit.git#subdirectory=clio-kit-mcp-servers/ndp",
|
| 109 |
+
"ndp-mcp",
|
| 110 |
+
])
|
| 111 |
+
async with stdio_client(params) as (r, w):
|
| 112 |
+
async with ClientSession(r, w) as s:
|
| 113 |
+
await s.initialize()
|
| 114 |
+
out = await s.call_tool(name, args)
|
| 115 |
+
print("".join(c.text for c in out.content if hasattr(c, "text")))
|
| 116 |
+
|
| 117 |
+
asyncio.run(call())
|
| 118 |
+
```
|
| 119 |
+
|
| 120 |
+
Save as `test.py` and run:
|
| 121 |
+
|
| 122 |
+
```bash
|
| 123 |
+
uv run --isolated test.py
|
| 124 |
+
```
|
| 125 |
+
|
| 126 |
## Files
|
| 127 |
|
| 128 |
- `merged_16bit/` — full safetensors checkpoint
|