Havenlon/ Execution-World-Qwen35-A3B-Q4_K_M:V1

69 2 months ago

A Havenlon-focused Qwen AgentWorld 35B-A3B model for local agent-environment simulation, next-state prediction, execution consequence reasoning, and controlled real-world AI research. 

ollama run Havenlon/Execution-World-Qwen35-A3B-Q4_K_M:V1

Details

2 months ago

88aac3c39567 · 21GB

model
qwen35moe
·
34.7B
·
Q4_K_M
system
你是一个语言世界模型，负责模拟 Agent 动作之后的环境反馈。 根据已有环
params
{ "num_ctx": 32768, "num_predict": 2048, "temperature": 0.6, "top_k": 20, "top_p

Readme

Execution-World-Qwen35-A3B-Q4_K_M

A Havenlon-focused Qwen AgentWorld 35B-A3B model for reasoning about environments, next-state transitions, execution consequences, and how AI decisions propagate toward real-world outcomes.

Execution World is complementary to the Havenlon Execution Boundary model family.

Focus

  • Agent-environment simulation
  • Next-state prediction
  • Action consequence reasoning
  • World-state transitions
  • Multi-step agent behavior
  • Execution consequence analysis
  • Real-world action simulation
  • Havenlon research on controlled AI execution

Run

ollama run Havenlon/Execution-World-Qwen35-A3B-Q4_K_M:V1

Quantization

Q4_K_M

Boundary and World

Havenlon separates two different problems.

Execution Boundary

The Boundary model asks:

Should this action be allowed to happen?

It focuses on policy, evidence, intent, authorization, execution constraints, and independent veto.

Execution World

The World model asks:

If this action happens, what state is likely to follow?

It focuses on environment dynamics, action consequences, next-state prediction, and multi-step interaction.

The two roles are intentionally different:

Boundary decides. World simulates.

Model scale follows the task rather than the other way around.

Execution World is intended as a research and simulation layer for understanding how AI decisions propagate through environments before those decisions are allowed to become real-world execution.