SkyIsNotGreen/hive-tool-manager-1.7b-v1

Hive Tool Manager 1.7B

The Hive 1.7B roster

This model is one seat in the five-model Hive 1.7B family. They are role specialists, not five general chat models:

model	seats
Hive Conductor 1.7B	controller + planner
Hive Reader 1.7B	mechanics + RAG
Hive Auditor 1.7B	placement + critic
Hive Dispatcher 1.7B	dispatcher
Hive Tool Manager 1.7B	agency

The family is served and routed by Hive.

A tool manager for the Hive local control plane: choose the right tool and sequence tool calls.

Built from LiquidAI/LFM3-1.7B-Base, this specialist covers the agency seat. It is not a general chat model.

What it is

• The agency seat: choose a tool for a job, or return an ordered multi-tool plan.
• Tool set: read_file, write_file, edit_file, bash, run_tests, web_search, image_gen.
• Two task families: single-tool choice and multi-tool ordering.
• Evaluated on 43 held-out agency tasks; see the table below.

Use cases

• Coding-agent tool choice. Pick read_file, write_file, edit_file, bash or run_tests for the next step, instead of guessing from a generic prompt.
• Test-aware edit loops. Return read_file -> edit_file -> run_tests for a change that must be inspected, edited and verified.
• Local automation. Choose bash for a shell job, web_search for current information, image_gen for a visual asset, and sequence the calls when more than one is needed.
• Research-to-write flows. Order web_search -> read_file -> write_file when a task needs sources, a local draft and a saved result.
• Tool-plan validation in the harness. Use the ordered tool list as the agency step before a local agent executes anything.

Tool Manager covers tool choice and ordering. It does not execute the tools or act as a general chat model.

Evaluation

Held-out Hive role suites, greedy decoding:

role	path	accuracy	n	median
agency	HF	0.977	43	0.036 s
agency	GGUF	0.930	43	0.048 s

Reproduce

The held-out task files, deterministic checkers and endpoint evaluator are in the Hive harness. The JSON files under eval/ are the exact evaluation records: greedy decoding, response_format enabled for JSON-shaped tasks, and Q4_K_M for the GGUF row. Weight hashes are in SHA256SUMS.

Training

• Base: LiquidAI/LFM3-1.7B-Base
• Method: TRL SFT + LoRA (r=16, alpha=32, dropout 0.05)
• Data: synthetic, checker-validated Hive role rows; no user data
• Format: LFM3 chat template with enable_thinking=false; assistant-only loss
• Run: 360 rows, 5.0 epochs, 115 steps
• Peak VRAM: 4.4 GiB on one 20 GB AMD card

Usage

Merged transformers model:

from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "SkyIsNotGreen/hive-tool-manager-1.7b-v1"
model = AutoModelForCausalLM.from_pretrained(
    repo, subfolder="model", dtype="bfloat16", trust_remote_code=True)
tok = AutoTokenizer.from_pretrained(
    repo, subfolder="model", trust_remote_code=True)

The merged model is the supported checkpoint; this release does not ship a standalone adapter.

GGUF with llama.cpp:

llama-server -m gguf/hive-tool-manager-1.7b-v1-Q4_K_M.gguf --alias tool-manager-1.7b \
  --host 127.0.0.1 --port 8164 -ngl 999 -c 8192 -np 2 \
  --jinja --reasoning off --no-webui

--reasoning off is required for this LFM3 template.

Run with Hive

hive serve can fetch this GGUF and start it as a role tier; see the Hive install guide. The direct llama-server command above is the minimal portable path.

Limitations

Agency-only release split from Operator v2. It handles tool selection and ordering, not general chat.

License and attribution

This is a derivative of LiquidAI/LFM3-1.7B-Base, released under the LFM Open License v1.0 (see LICENSE). The base model is by Liquid AI. Commercial use by an entity above USD 10M annual revenue is not licensed under LFM 1.0.

Modified files: merged weights and GGUF conversion. See NOTICE.