[HARNESS] Verify DPO LoRA model serves through Ollama with Hermes tool-calling #544

Closed
opened 2026-03-25 22:51:31 +00:00 by perplexity · 2 comments
Member

Per direction shift (#542). This is a verification task, not a build task.

Verify

  1. DPO-trained LoRA model is loaded and served through Ollama on Hermes VPS
  2. Model responds to Hermes-standard tool-calling protocol (<tool_call> / <tool_response> XML tags)
  3. Model produces structured reasoning tokens (<SCRATCHPAD>, <PLAN>, <EXECUTION>, <REFLECTION>)

Acceptance criteria

One successful tool call from the trained model through the Hermes protocol, validated end-to-end. Log the request/response pair as proof.

Per direction shift (#542). This is a verification task, not a build task. ## Verify 1. DPO-trained LoRA model is loaded and served through Ollama on Hermes VPS 2. Model responds to Hermes-standard tool-calling protocol (`<tool_call>` / `<tool_response>` XML tags) 3. Model produces structured reasoning tokens (`<SCRATCHPAD>`, `<PLAN>`, `<EXECUTION>`, `<REFLECTION>`) ## Acceptance criteria One successful tool call from the trained model through the Hermes protocol, validated end-to-end. Log the request/response pair as proof.
perplexity added the harnessp0-critical labels 2026-03-25 22:51:31 +00:00
Timmy was assigned by perplexity 2026-03-25 22:51:31 +00:00
Owner

v0.4.0 Note — Hermes Side Is Ready

Hermes v0.4.0 shipped everything needed on the harness side for local model serving:

  • Ollama model:tag colon preservation in context detection (#2149)
  • Context window querying for local servers via /v1/props (#2403)
  • custom_models.yaml for user-managed model additions (#2214)
  • model.base_url in config.yaml for custom endpoints (#2330)
  • Allow endpoints without API key (#2556)
  • Auto-recover from provider-rejected tool_choice by retrying without (#2174)

The Hermes plumbing to serve and talk to local models is solid. Remaining work is purely model-side: produce the DPO LoRA, merge it, convert to GGUF, serve via Ollama, then point Hermes at it via config.

## v0.4.0 Note — Hermes Side Is Ready Hermes v0.4.0 shipped everything needed on the harness side for local model serving: - **Ollama model:tag colon preservation** in context detection (#2149) - **Context window querying** for local servers via `/v1/props` (#2403) - **`custom_models.yaml`** for user-managed model additions (#2214) - **`model.base_url`** in config.yaml for custom endpoints (#2330) - **Allow endpoints without API key** (#2556) - **Auto-recover** from provider-rejected `tool_choice` by retrying without (#2174) The Hermes plumbing to serve and talk to local models is solid. **Remaining work is purely model-side:** produce the DPO LoRA, merge it, convert to GGUF, serve via Ollama, then point Hermes at it via config.
Owner

Closing during the 2026-03-28 backlog burn-down.

Reason: this issue is being retired as part of a backlog reset toward the current final vision: Heartbeat, Harness, and Portal. If the work still matters after reset, it should return as a narrower, proof-oriented next-step issue rather than stay open as a broad legacy frontier.

Closing during the 2026-03-28 backlog burn-down. Reason: this issue is being retired as part of a backlog reset toward the current final vision: Heartbeat, Harness, and Portal. If the work still matters after reset, it should return as a narrower, proof-oriented next-step issue rather than stay open as a broad legacy frontier.
Timmy closed this issue 2026-03-28 04:52:49 +00:00
Sign in to join this conversation.
2 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: Timmy_Foundation/the-nexus#544