Yantra
1B AgentMITSub-1B Agentic Function-Calling Model
Yantra 1B Agent is a sub-1b agentic function-calling model. Operates across Parameters: 1B (4-bit GGUF) · Base: MiniCPM5-1B · Method: DTSA + Router V2 · Task Success: 3.52× baseline · Training: ToolACE benchmark · Format: GGUF · 4-bit · Runtime: llama.cpp · Colab · License: MIT. License: MIT.
A sub-1B open-weight agentic model (MiniCPM5-1B) fine-tuned on ToolACE with decoupled task routing (DTSA + Router V2). Achieves 3.52× baseline task success on agentic benchmarks. Quantized to 4-bit GGUF for on-device deployment. MIT license. Runs in Colab, llama.cpp, and any GGUF-compatible runtime.
- ✓3.52× baseline on ToolACE agentic benchmarks
- ✓Sub-1B parameters — runs on any consumer hardware
- ✓4-bit quantized GGUF for zero-GPU deployment
- ✓690+ downloads in first weeks of release
- ✓Decoupled routing: separate task selector from executor
Get started in minutes
# Use with llama.cpp or Ollama
ollama pull eulogik/yantra-1b-agent# Python with llama-cpp-python
from llama_cpp import Llama
llm = Llama.from_pretrained(
repo_id="eulogik/yantra-1b-agent",
filename="yantra-1b-q4_k_m.gguf",
)
response = llm.create_chat_completion(
messages=[{"role": "user", "content": "Search for today's weather in Mumbai"}],
tools=[{"type": "function", "function": {"name": "search", "parameters": {}}}],
)
print(response["choices"][0]["message"])Cite this Model & Architecture
If you utilize Yantra in your academic research, benchmarks, or enterprise deployments, please cite our open-weight publication:
@misc{eulogik2026yantra1bagent,
title = {Yantra 1B Agent: Sub-1B Agentic Function-Calling Model},
author = {Kishore, Gautam and Eulogik Systems Engineering},
year = {2026},
publisher = {Eulogik},
howpublished = {\url{https://eulogik.com/models/yantra-1b-agent}}
}Need help deploying Yantra?
We offer consulting, custom fine-tuning, and on-premise deployment for all our models.