Instructions to use ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16")
model = AutoModelForCausalLM.from_pretrained("ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

PEFT
How to use ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16 with PEFT:
```
Task type is invalid.
```
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16

SGLang

How to use ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Unsloth Studio

How to use ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16 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 ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16 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 ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16 to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16 to start chatting

Load model with FastModel

pip install unsloth
from unsloth import FastModel
model, tokenizer = FastModel.from_pretrained(
    model_name="ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16",
    max_seq_length=2048,
)

Docker Model Runner
How to use ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16 with Docker Model Runner:
```
docker model run hf.co/ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16
```

MiniCPM5-1B Agentic Tooluse Merged FP16

Latest Status: July 2026 Nemotron Repair

This is the current best merged fp16 release for the MiniCPM5-1B Agentic Tooluse family. It was repaired with a short Nemotron SFT continuation followed by DPO preference optimization, starting from the previously trained DPO adapter.

External Team-ACE/ToolACE evaluation, 300 cases:

Metric	Base MiniCPM5-1B	Merged fp16	Delta
parseable_rate	0.0133	0.9933	+0.9800
valid_name_rate	0.0133	0.9700	+0.9567
expected_name_rate	0.0133	0.9267	+0.9133
args_exact_rate	0.1500	0.6533	+0.5033
arg_key_overlap	0.0033	0.7517	+0.7484
no_schema_copy_rate	1.0000	1.0000	+0.0000
no_repetition_rate	0.9967	1.0000	+0.0033
stopped_cleanly_rate	0.0000	0.1500	+0.1500

Use this adapter for first-call XML tool selection. In production, stop generation immediately after the first complete </function> call.

Standalone merged fp16 Hugging Face model built from openbmb/MiniCPM5-1B plus the MiniCPM5-1B-Agentic-Tooluse-QLoRA-v2 adapter.

Use this repository when you want a complete Transformers/vLLM-loadable model instead of loading a separate PEFT adapter.

Model Family

Use case	Repository
Standalone fp16 Hugging Face model for Transformers or vLLM	This repo: `ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16`
Adapter-only PEFT/LoRA loading	`ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-QLoRA-v2`
Quantized GGUF files for llama.cpp-compatible runtimes	`ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-GGUF`

Quick Answer

This is the current merged fp16 release of a MiniCPM5-1B tool-calling model after the July 2026 Nemotron SFT+DPO repair. It is optimized for first-call function selection in MiniCPM5 XML format and should be used with a runtime that stops generation after the first completed function call.

Model Details

Base model: openbmb/MiniCPM5-1B
Adapter source: ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-QLoRA-v2
Release type: merged fp16 model
File format: Hugging Face safetensors
Primary task: tool calling, function calling, XML function-call generation
Recommended runtime behavior: stop after first completed </function>

Tool Calling Format

MiniCPM5 uses XML-style function calls:

<function name="tool_name"><param name="param_name">value</param></function>

Use the tokenizer chat template and pass tools through the template when possible. Decode with skip_special_tokens=False for XML parsing and evaluation.

Transformers Example

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16"

tokenizer = AutoTokenizer.from_pretrained("openbmb/MiniCPM5-1B", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True,
)
model.eval()

vLLM

Use this repo as the model path:

vllm serve ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16 --trust-remote-code

MiniCPM-family support depends on the installed vLLM version and parser support available in that build.

Evaluation

External ToolACE-Derived Sanity Eval

External sanity eval from Team-ACE/ToolACE, not used in training. This is not the official ToolACE leaderboard evaluator.

n=300, greedy decoding, runtime stops after first completed function call.

Metric	Base MiniCPM5-1B	Fine-tuned merged model	Delta
parseable_rate	0.3567	0.9967	+0.6400
valid_name_rate	0.3467	0.9400	+0.5933
expected_name_rate	0.3467	0.8400	+0.4933
no_schema_copy_rate	0.3567	0.9967	+0.6400
no_repetition_rate	1.0000	1.0000	+0.0000

Held-Out Source-Mix Eval

Held-out split from the same xLAM/Glaive source mixture used for training. This is not an external benchmark.

Metric	Score
parseable_rate	0.7800
valid_name_rate	0.7733
expected_name_rate	0.7633
no_schema_copy_rate	0.7800
no_repetition_rate	0.7800

Training Data and Caveat

The adapter used to create this merged model was trained from:

A later audit found that the original validation checked for valid tool calls but did not strictly enforce that the supervised target ended immediately after the first tool call. Some source examples include assistant text, tool responses, follow-up turns, or final assistant answers. This release should therefore be framed as a first-call tool-selection model, not a clean-stop free-running agent.

Limitations

May continue with extra text or synthetic turns if generation is not stopped at </function>.
External failures are often near-miss function names, casing differences, or unusual tool names.
Not an official ToolACE/BFCL leaderboard score.
Not validated for production autonomous tool execution.
Quantized GGUF behavior may differ from this fp16 source model.

Safety

Tool calls should be parsed, validated, allowlisted, and executed only inside a controlled tool runtime. Do not directly execute generated tool calls for sensitive or irreversible actions.

Search Keywords

MiniCPM5, MiniCPM5-1B, OpenBMB, agentic tool calling, function calling, tool use LLM, XML function call, vLLM MiniCPM, Transformers safetensors, merged LoRA, QLoRA merged model, PEFT merged model, llama.cpp conversion source, Unsloth compatible model.

July 2026 Nemotron SFT+DPO Repair Update

This merged fp16 Transformers/vLLM release now reflects the latest repair pass from the MiniCPM5-1B Agentic Tooluse model family.

The repair continued from the previous DPO adapter and added a targeted Nemotron SFT + DPO pass focused on:

first-call tool selection
valid function-name prediction
argument-key selection
schema-copy avoidance
repeated-call suppression
XML <function ...>...</function> tool-call formatting

Training/eval artifacts:

SFT continuation: ./minicpm5_tooluse_nemotron_sft_from_repair_dpo
DPO continuation: ./minicpm5_tooluse_nemotron_dpo_from_repair_dpo
Merged fp16 export: ./minicpm5_tooluse_nemotron_merged_fp16
External eval: Team-ACE/ToolACE, 300 cases

Datasets used in the repair:

nvidia/Nemotron-SFT-Agentic-v2
nvidia/Nemotron-RL-Agentic-Function-Calling-Pivot-v1
nvidia/Nemotron-RL-Agentic-Conversational-Tool-Use-Pivot-v1

Dataset inspection note: nvidia/Nemotron-SFT-Agentic-v2/data/tool_calling.jsonl contains 8,444 physical lines and one malformed JSONL line. The final training pipeline reads the file directly with hf_hub_download, parses JSONL line by line, and skips invalid rows. The pipeline also removes oversized source policy text from SFT prompts while preserving tool schemas and recent dialogue context.

Latest External ToolACE Evaluation

Metric	Base MiniCPM5-1B	Nemotron repaired adapter/model	Delta
parseable_rate	0.0133	0.9933	+0.9800
valid_name_rate	0.0133	0.9700	+0.9567
expected_name_rate	0.0133	0.9267	+0.9133
args_exact_rate	0.1500	0.6533	+0.5033
arg_key_overlap	0.0033	0.7517	+0.7484
no_schema_copy_rate	1.0000	1.0000	+0.0000
no_repetition_rate	0.9967	1.0000	+0.0033
stopped_cleanly_rate	0.0000	0.1500	+0.1500

The repaired model beats base on 7/8 external ToolACE metrics.

The strongest improvements are in parseability, valid tool-name selection, expected tool-name selection, and argument-key overlap. The remaining weak point is clean stopping, so production runtimes should stop generation immediately after the first complete </function> call.

Runtime Contract

Use this as a first-call tool-selection model:

Render tools in the prompt.
Generate one tool call.
Parse the first complete <function ...>...</function> block.
Stop at </function>.
Execute the tool outside the model.
Feed the tool result back as the next turn.

Do not let the model keep generating fake tool responses or synthetic user turns after a function call.

Current Files vs Previous Versions

This repository keeps old versions in Hugging Face commit history. Files removed from the current visible file list are not permanently erased; they can still be found by opening the repository History tab and selecting an older commit.

Current merged fp16 weights are stored as sharded safetensors:

model-00001-of-00002.safetensors
model-00002-of-00002.safetensors
model.safetensors.index.json

The older single-file model.safetensors was removed from the current file list because it was from the previous merged release and could confuse loaders. It is still recoverable from the repository commit history.

Current Status

This is the current merged fp16 release after the July 2026 Nemotron SFT+DPO repair. Use it when you want a standalone Transformers/vLLM-loadable model without separately loading PEFT adapters.

Current primary weight files:

model-00001-of-00002.safetensors
model-00002-of-00002.safetensors
model.safetensors.index.json

The older single-file model.safetensors from the previous merged release was removed from the current file list because it conflicted with the new sharded weights. It remains recoverable from Hugging Face commit history.

Known Limitations

This is a merged fp16 model, so it does not require loading a separate PEFT adapter.
Strongest validated use case is first-call XML tool selection with Transformers/vLLM-style serving.
External ToolACE eval improved strongly after the July 2026 Nemotron SFT+DPO repair, but this is still not an official ToolACE/BFCL leaderboard submission.
Argument exactness is improved but not perfect; production systems should validate tool arguments before execution.
Clean stopping is still runtime-sensitive. Use stop strings or parser-side stopping after the first complete </function> block.
Not validated as a fully autonomous production agent. Use it inside an application-owned tool loop.

Downloads last month: -

Safetensors

Model size

1B params

Tensor type

F16

Model tree for ewinregirgojr/MiniCPM5-1B-Agentic-Tooluse-Merged-FP16

Base model

openbmb/MiniCPM5-1B

Finetuned

(33)

this model

Evaluation results

Parseable tool call rate on External ToolACE-derived first-call sanity eval
self-reported

0.993
Expected tool name rate on External ToolACE-derived first-call sanity eval
self-reported

0.927