Open
Conversation
Split inference/src/ into inference/core/, server/, cli/, shaders/. Move tests/ into inference/tests/. Rewrite server.m for OpenAI API conformance: - Proper SSE chunks with model, created, system_fingerprint, role delta - Non-streaming mode (stream: false) - Tool-calling: parse tools array, detect <tool_call> blocks, emit as OpenAI tool_calls in both streaming and non-streaming responses - Tool-result round-trips: handle tool role messages and assistant tool_calls in conversation history - /v1/models lists all three models (35B-A3B, 27B, 9B) with loaded indicator - /health returns model metadata, uptime, engine config - Usage (prompt_tokens, completion_tokens) and timing (prefill_ms, decode_ms, tokens_per_sec) in final chunk and non-streaming response - OpenAI-format error objects for 400/404/500 Add scripts/smoke_test.sh: build, server API conformance, tok/s benchmark and quality check for all models present. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Create harness/ Rust crate (tokio, axum, serde) with the core type system for multi-model orchestration: - roles: 8 canonical roles (Planner, TaskManager, Worker, Reviewer, Referee, Tiebreaker, Summarizer, Searcher) with scratchpad access rules per role - tasks: TaskSpec, TaskResult, TaskGrade (dimensional scores), TaskTelemetry, TaskState lifecycle, Task aggregate - tasks/scratchpad: per-task directory + append-only manifest - bus: Envelope (hub-routed, task_id as single source of truth, from_role/to_role on envelope), Participant (identity only), 9 structured Message variants - providers: async Provider trait (chat, chat_stream via Pin<Box<dyn Stream>>, load_model, unload_model, health_check), normalized ChatRequest/ChatResponse/StreamEvent types - scheduler: LocalCapacity, CloudCapacity, LoadedModel, SchedulerDecision types - tools: ToolName, ToolSpec, ToolResult - config: HarnessConfig, RoleBinding, ProviderConfig - telemetry: ModelPerformanceRecord - types: DurationMs, TimestampMs shared newtypes Add Rust target/ to .gitignore. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
… dispatch The harness now has two provider interfaces reflecting two fundamentally different relationships: - InferenceProvider: turn-level, harness drives the tool loop (orome-local) - CliProvider: task-level, CLI drives its own tool loop (claude, codex) Adds plan file schema (.orome/plan.yaml) and boot manifest (orome.yaml) as the contracts between planning and execution phases. Config updated to reflect CLI-based providers (no API keys, no hardcoded capacity). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…plates Implements the full dispatch pipeline: - Dispatcher routes tasks to CLI providers (claude/codex) or local inference (orome), driving the tool loop for local models - `orome plan --claude` launches interactive planning with injected system prompt (project context, schemas, model roster) - `orome run --claude` reads .orome/plan.yaml, validates the DAG, and executes the worker -> reviewer -> retry loop - Provider adapters: claude (JSON output), codex (JSONL), orome-local (OpenAI-compatible HTTP/SSE with x_orome timing) - Prompt templates for planner, worker, reviewer, and tiebreaker roles Dry-run validates plan parsing and topological sort work correctly. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.