Files

jgrusewski 12fdd18223 refactor: remove entire CPU training path — 5,307 lines of dead code

Deleted:
- DQN::compute_loss_internal (280 lines) — old Candle forward+loss
- DQN::train_step (55 lines) — old Candle training step
- DQN::compute_gradients (47 lines) — old gradient accumulation
- ComputeLossResult struct — only used by deleted functions
- RegimeConditionalDQN::train_step (65 lines) — old dispatch
- RegimeConditionalDQN::train_step_gpu_regime (100 lines) — old GPU path
- RegimeConditionalDQN::compute_gradients_gpu (130 lines) — old regime gradients
- RegimeConditionalDQN::compute_gradients (92 lines) — old dispatch
- DQNAgentType::train_step dispatch — dead
- DQNAgentType::compute_gradients dispatch — dead
- GpuDqnTrainer::upload_batch (71 lines) — old CPU→GPU upload
- train_step.rs (500 lines) — entire module including ensure_fused_ctx
- dqn_benchmark.rs — used old train_step
- examples.rs — used old train_step
- validation/adapters.rs (289 lines) — used old train_step
- dqn/trainable_adapter.rs — used old train_step
- gpu_smoketest.rs — tested old train_step
- Gradient accumulation path in training_loop.rs (144 lines)
- IQN d_h_s2().clone() → raw pointer (zero alloc)
- Causal intervention format! string alloc removed
- Dead HER relabel functions (320 lines)

Kept:
- ensure_fused_ctx logic inlined into training_loop.rs
- set_noise_sigma_scale re-added to RegimeConditionalDQN

Fixed:
- GpuReplayBuffer max_batch_size wired from batch_size parameter
  (was hardcoded 1024, blocking batch_size=8192)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

2026-04-02 09:06:23 +02:00

2.3 KiB

Raw Permalink Blame History

name, description

name	description
sparc-integration	🔗 System Integrator - You merge the outputs of all modes into a working, tested, production-ready system. You ensure co...

🔗 System Integrator

Role Definition

You merge the outputs of all modes into a working, tested, production-ready system. You ensure consistency, cohesion, and modularity.

Custom Instructions

Verify interface compatibility, shared modules, and env config standards. Split integration logic across domains as needed. Use new_task for preflight testing or conflict resolution. End integration tasks with attempt_completion summary of what's been connected.

Available Tools

read: File reading and viewing
edit: File modification and creation
browser: Web browsing capabilities
mcp: Model Context Protocol tools
command: Command execution

Usage

Option 1: Using MCP Tools (Preferred in Claude Code)

mcp__claude-flow__sparc_mode {
  mode: "integration",
  task_description: "connect payment service",
  options: {
    namespace: "integration",
    non_interactive: false
  }
}

Option 2: Using NPX CLI (Fallback when MCP not available)

# Use when running from terminal or MCP tools unavailable
npx claude-flow sparc run integration "connect payment service"

# For alpha features
npx claude-flow@alpha sparc run integration "connect payment service"

# With namespace
npx claude-flow sparc run integration "your task" --namespace integration

# Non-interactive mode
npx claude-flow sparc run integration "your task" --non-interactive

Option 3: Local Installation

# If claude-flow is installed locally
./claude-flow sparc run integration "connect payment service"

Memory Integration

Using MCP Tools (Preferred)

// Store mode-specific context
mcp__claude-flow__memory_usage {
  action: "store",
  key: "integration_context",
  value: "important decisions",
  namespace: "integration"
}

// Query previous work
mcp__claude-flow__memory_search {
  pattern: "integration",
  namespace: "integration",
  limit: 5
}

Using NPX CLI (Fallback)

# Store mode-specific context
npx claude-flow memory store "integration_context" "important decisions" --namespace integration

# Query previous work
npx claude-flow memory query "integration" --limit 5

2.3 KiB Raw Permalink Blame History