From 30f8ca38630fc209c1bf03e11f6674604f03db98 Mon Sep 17 00:00:00 2001 From: Hermes Agent Date: Tue, 28 Apr 2026 17:19:45 +0000 Subject: [PATCH 1/3] Add AI model optimizer cron job draft and initial state files --- assets/ai-optimizer/CRON_EXECUTION_PROMPT.md | 203 +++++++++++++ assets/ai-optimizer/CRON_JOB_DRAFT.md | 283 +++++++++++++++++++ assets/ai-optimizer/results.csv | 1 + assets/ai-optimizer/state.json | 21 ++ 4 files changed, 508 insertions(+) create mode 100644 assets/ai-optimizer/CRON_EXECUTION_PROMPT.md create mode 100644 assets/ai-optimizer/CRON_JOB_DRAFT.md create mode 100644 assets/ai-optimizer/results.csv create mode 100644 assets/ai-optimizer/state.json diff --git a/assets/ai-optimizer/CRON_EXECUTION_PROMPT.md b/assets/ai-optimizer/CRON_EXECUTION_PROMPT.md new file mode 100644 index 0000000..20a6600 --- /dev/null +++ b/assets/ai-optimizer/CRON_EXECUTION_PROMPT.md @@ -0,0 +1,203 @@ +# AI Model Optimization Cron Job - EXECUTION PROMPT + +**When this cron runs, follow these instructions exactly:** + +--- + +## Your Role + +You are an AI model optimization agent. Your task is to find the best ollama/llama.cpp configuration for maximum context size and hardware utilization. + +**Hardware:** +- 2× AMD MI50 GPUs (32GB VRAM each, 64GB total) +- 128GB system RAM +- ROCm: HSA_OVERRIDE_GFX_VERSION=9.0.6, HIP_VISIBLE_DEVICES=0,1 + +--- + +## File Locations + +``` +STATE: /opt/data/infra/assets/ai-optimizer/state.json +RESULTS: /opt/data/infra/assets/ai-optimizer/results.csv +INFRA_REPO: /opt/data/infra +``` + +--- + +## Model Queues + +### GPU Track (Coding - prioritize speed + context on GPU) +1. `devstral-small-2:24b` +2. `qwen2.5-coder:32b` +3. `codellama:34b-instruct` + +### RAM Track (Knowledge - prioritize max context) +1. `qwen2.5:72b` +2. `nemotron-3-nano:30b` +3. `mixtral:8x7b-instruct` + +--- + +## Context Steps (in order) +``` +[32768, 65536, 98304, 131072, 163840, 200704, 262144, 327680] +``` + +--- + +## Each Run - Step by Step + +### 1. Read State +```bash +cd /opt/data/infra +cat assets/ai-optimizer/state.json +``` + +### 2. Determine Next Test +- Read `track` (gpu or ram) +- Read `current_model` from queue at `model_index` +- Read `current_config` for parameters to test +- Select next context step from `context_steps` based on `phase` + +### 3. Pull Model (if needed) +```bash +docker exec ollama ollama list | grep -q "" || docker exec ollama ollama pull +``` + +### 4. Create Test Modelfile +```bash +docker exec ollama bash -c "cat < /root/.ollama/test_${model}.modelfile +FROM ${model} +PARAMETER num_ctx ${current_config.num_ctx} +PARAMETER num_gpu ${current_config.num_gpu} +PARAMETER flash_attn ${current_config.flash_attn} +PARAMETER num_predict 4096 +PARAMETER num_keep 1024 +PARAMETER repeat_penalty 1.1 +EOF" + +docker exec ollama ollama create test-model -f /root/.ollama/test_${model}.modelfile +``` + +### 5. Run Benchmark +```bash +# Warm up +docker exec ollama ollama run test-model "Hello" > /dev/null + +# Coding prompt +START=$(date +%s%N) +docker exec ollama ollama run test-model "Write a Python async context manager that retries a function with exponential backoff, max 5 retries, and logs each attempt using structlog. Include type hints." +END=$(date +%s%N) + +# Calculate tokens/sec from output +``` + +### 6. Measure VRAM (if possible) +```bash +# Try host first +rocm-smi --showmeminfo vram 2>/dev/null || \ +# Try via docker +docker exec --privileged ollama rocm-smi --showmeminfo vram 2>/dev/null || \ +# Fallback +echo "VRAM measurement unavailable" +``` + +### 7. Record Results +- Parse tokens/sec from ollama output +- Record VRAM/RAM usage +- Determine if this is best config so far for this model +- Update `best_configs` if tokens/sec improved or context increased + +### 8. Update State +```python +# Logic: +if test_successful: + if context_step < max_reached: + phase = "context_scaling" + current_config.num_ctx = next_context_step + else: + # Move to next model + model_index += 1 + phase = "context_scaling" + current_config.num_ctx = context_steps[0] +else: + # OOM or error - record last good as best + best_configs[track][current_model] = last_good_config + model_index += 1 + phase = "context_scaling" +``` + +### 9. Commit to Repo +```bash +cd /opt/data/infra +git add assets/ai-optimizer/ +git commit -m "ai-optimizer: tested ${model} at ${num_ctx} ctx - ${status}" +git push origin master +``` + +### 10. Matrix Notification (if available) +```python +import os +if os.getenv("MATRIX_HOME_SERVER") and os.getenv("MATRIX_ACCESS_TOKEN"): + # Send notification to Matrix room + # Room ID from env or config + pass +# Else: silent +``` + +--- + +## Stop Conditions + +1. All models in both queues have `best_configs` recorded +2. Manual intervention needed (error in state.json `error` field) +3. No progress for 3 consecutive runs (stuck) + +--- + +## Error Handling + +If any step fails: +1. Log error to state.json: `"error": {"message": "...", "timestamp": "..."}` +2. Do NOT increment model_index (retry next run) +3. Commit state with error field +4. Exit gracefully + +--- + +## Important Notes + +- **No num_parallel**: Do not use this parameter +- **Two tracks**: Complete GPU track first, then RAM track +- **Backend**: Start with ollama, llama.cpp testing is optional (requires uncommenting in compose.yml) +- **Host access**: Some commands need host - use docker exec or SSH if available +- **Ask before deploy**: If config changes needed in NixOS modules, show diff and wait for user confirmation before `nh os switch` + +--- + +## Example State Transitions + +**Start:** +```json +{"track": "gpu", "model_index": 0, "current_model": "devstral-small-2:24b", "current_config": {"num_ctx": 32768, ...}} +``` + +**After successful test at 32k:** +```json +{"track": "gpu", "model_index": 0, "current_model": "devstral-small-2:24b", "current_config": {"num_ctx": 65536, ...}} +``` + +**After OOM at 131k:** +```json +{ + "track": "gpu", + "model_index": 1, + "current_model": "qwen2.5-coder:32b", + "best_configs": { + "gpu": { + "devstral-small-2:24b": {"num_ctx": 98304, "num_gpu": 99, "tokens_per_sec": 11.2} + } + } +} +``` diff --git a/assets/ai-optimizer/CRON_JOB_DRAFT.md b/assets/ai-optimizer/CRON_JOB_DRAFT.md new file mode 100644 index 0000000..3ff0781 --- /dev/null +++ b/assets/ai-optimizer/CRON_JOB_DRAFT.md @@ -0,0 +1,283 @@ +# AI Model Optimization Cron Job + +**Goal:** Find optimal configurations for maximum context size with full hardware utilization. + +**Hardware:** +- 2× AMD MI50 GPUs (32GB VRAM each, 64GB total) +- 128GB system RAM +- ROCm: HSA_OVERRIDE_GFX_VERSION=9.0.6, HIP_VISIBLE_DEVICES=0,1 + +--- + +## Model Queue + +### GPU-Optimized (Coding - prioritize speed + context on GPU) +1. `devstral-small-2:24b` - Best coding model +2. `qwen2.5-coder:32b` - Strong coder, fits on GPU+offload +3. `codellama:34b-instruct` - Legacy but solid + +### RAM-Optimized (Knowledge - prioritize max context, accept slower) +1. `qwen2.5:72b` - Best knowledge, needs heavy offload +2. `nemotron-3-nano:30b` - Good general knowledge +3. `mixtral:8x7b-instruct` - MoE, efficient for knowledge + +--- + +## Optimization Strategy + +**Two separate tracks:** + +### Track A: GPU-Focused (Coding) +``` +Baseline: num_ctx=32768, num_gpu=99, flash_attn=true +Steps: +1. Increase context: 32k → 65k → 98k → 131k → 163k +2. At each step, verify VRAM usage < 60GB (leave headroom) +3. If OOM: reduce num_gpu until stable, record best +4. Measure tokens/sec - if < 5 tok/s, consider context too high +``` + +### Track B: RAM-Focused (Knowledge) +``` +Baseline: num_ctx=65536, num_gpu=50, flash_attn=true +Steps: +1. Increase context: 65k → 131k → 200k → 262k → 327k +2. Allow heavy RAM offload (system RAM up to 100GB) +3. If OOM: reduce context or num_gpu +4. Speed less critical - focus on max stable context +``` + +--- + +## Backend-Specific Configs + +### Ollama (Modelfile parameters) +``` +PARAMETER num_ctx +PARAMETER num_gpu +PARAMETER flash_attn true/false +PARAMETER num_predict 4096 +PARAMETER num_keep 1024 +PARAMETER repeat_penalty 1.1 +``` + +### Llama.cpp (CLI flags) +``` +--ctx-size +--n-gpu-layers +--flash-attn on/off +--n-predict 4096 +--batch-size 4096 +--ubatch-size 512 +--cache-type-k f16 +--cache-type-v f16 +--split-mode layer +--no-mmap +``` + +--- + +## Host Test Instructions + +**The cron runs inside the hermes container. Some tests require host access:** + +### 1. VRAM Monitoring (HOST) +```bash +# Run on host to check VRAM usage during/after benchmark +sudo rocm-smi --showmeminfo vram + +# Or via docker exec if rocm-smi available in container +docker exec --privileged ollama rocm-smi --showmeminfo vram +``` + +### 2. Running Ollama Benchmarks (CONTAINER) +```bash +# Pull model +docker exec ollama ollama pull + +# Create custom modelfile +docker exec ollama bash -c 'cat < /root/.ollama/test.modelfile +FROM +PARAMETER num_ctx 65536 +PARAMETER num_gpu 99 +PARAMETER flash_attn true +EOF' + +# Create model from modelfile +docker exec ollama ollama create test-model -f /root/.ollama/test.modelfile + +# Run benchmark (warm model first) +docker exec ollama ollama run test-model "Write a Python async context manager with exponential backoff" + +# Cleanup +docker exec ollama ollama rm test-model +``` + +### 3. Running Llama.cpp Benchmarks (CONTAINER - needs llama.cpp container) +```bash +# Uncomment llama_cpp_devstral in compose.yml first +# Then rebuild: sudo nh os switch --flake .#lazyworkhorse + +# Test via HTTP API +curl http://localhost:8300/v1/completions \ + -H "Content-Type: application/json" \ + -d '{ + "model": "devstral-2-small-llama_cpp", + "prompt": "Write a Python function", + "max_tokens": 100 + }' +``` + +### 4. Deploying Changes (HOST via ai-worker) +```bash +# After optimization, commit results +cd /home/ai-worker/infra +git add assets/ai-optimizer/ +git commit -m "ai-optimizer: new best config for " +git push + +# If config changes needed in ollama_init_custom_models.nix: +# 1. Edit the file +# 2. nixpkgs-fmt . +# 3. Show diff to user +# 4. Wait for confirmation +# 5. sudo nh os switch --flake .#lazyworkhorse +``` + +### 5. Accessing Host from Hermes Container +```bash +# SSH to host as ai-worker (key should be mounted) +ssh -i /path/to/key ai-worker@host.docker.internal + +# Or via docker socket if mounted +# (not recommended for security) +``` + +--- + +## Benchmark Prompts + +### Coding (Track A) +``` +"Write a Python async context manager that retries a function with exponential backoff, max 5 retries, and logs each attempt using structlog. Include type hints and error handling." +``` + +### Knowledge (Track B) +``` +"Explain the complete memory hierarchy in modern GPUs, from registers through L1/L2 caches to VRAM, and how data moves between them during matrix multiplication. Include bandwidth considerations for each level." +``` + +### Measurement +- Tokens per second (generation speed) +- Time to first token (latency) +- VRAM usage (via rocm-smi) +- System RAM usage (via free -h) +- Context success (did it complete without OOM?) + +--- + +## State File Structure + +`/opt/data/infra/assets/ai-optimizer/state.json` + +```json +{ + "track": "gpu", + "current_model": "devstral-small-2:24b", + "model_index": 0, + "phase": "context_scaling", + "backend": "ollama", + "current_config": { + "num_ctx": 65536, + "num_gpu": 99, + "flash_attn": true + }, + "best_configs": { + "gpu": { + "devstral-small-2:24b": { + "backend": "ollama", + "num_ctx": 131072, + "num_gpu": 99, + "flash_attn": true, + "tokens_per_sec": 12.5, + "vram_used_gb": 58.2, + "tested_at": "2026-04-28T17:00:00Z" + } + }, + "ram": {} + }, + "completed_models": [], + "gpu_queue": ["devstral-small-2:24b", "qwen2.5-coder:32b", "codellama:34b-instruct"], + "ram_queue": ["qwen2.5:72b", "nemotron-3-nano:30b", "mixtral:8x7b-instruct"] +} +``` + +--- + +## Results CSV + +`/opt/data/infra/assets/ai-optimizer/results.csv` + +```csv +timestamp,track,model,backend,phase,num_ctx,num_gpu,flash_attn,tokens_per_sec,vram_gb,ram_gb,status,is_best +2026-04-28T17:00:00Z,gpu,devstral-small-2:24b,ollama,context_scaling,65536,99,true,15.2,52.1,18.4,success,false +``` + +--- + +## Cron Job Flow + +``` +1. Read state.json +2. If both queues empty → STOP (all models tested) +3. Select next model from current track queue +4. Pull model if needed (docker exec ollama ollama pull) +5. Create Modelfile / llama.cpp config with current test params +6. Run benchmark (both prompts) +7. Measure: tokens/sec, VRAM (rocm-smi), RAM (free -h) +8. If successful: + - Increase context (next step) + - Update current_config in state +9. If OOM/error: + - Record last good config as best_configs[track][model] + - Move to next model in queue +10. Update state.json +11. Append to results.csv +12. Git commit + push to /opt/data/infra +13. Send Matrix notification if available, else silent +``` + +--- + +## Matrix Notification (Optional) + +```python +# If matrix credentials available in environment +if os.getenv("MATRIX_HOME_SERVER") and os.getenv("MATRIX_ACCESS_TOKEN"): + # Send completion notification + # Room: !ai-optimizer:lazyworkhorse.net (or similar) + pass +# Else: silent, just commit +``` + +--- + +## Files to Create + +``` +/opt/data/infra/assets/ai-optimizer/ +├── state.json # Current progress +├── results.csv # All test results +├── best_configs.json # Final best configs (human-readable) +└── CRON_JOB_DRAFT.md # This file +``` + +--- + +## Notes + +- **No num_parallel**: Removed to avoid limiting other settings +- **Two tracks**: GPU (coding/speed) vs RAM (knowledge/context) +- **Both backends**: Test ollama first, then llama.cpp if available +- **Host tests**: rocm-smi must run on host or privileged container +- **Deploy**: ai-worker has sudo for nh/nixos-rebuild, must ask user first diff --git a/assets/ai-optimizer/results.csv b/assets/ai-optimizer/results.csv new file mode 100644 index 0000000..7e25194 --- /dev/null +++ b/assets/ai-optimizer/results.csv @@ -0,0 +1 @@ +timestamp,track,model,backend,phase,num_ctx,num_gpu,flash_attn,tokens_per_sec,vram_gb,ram_gb,status,is_best diff --git a/assets/ai-optimizer/state.json b/assets/ai-optimizer/state.json new file mode 100644 index 0000000..fff69f9 --- /dev/null +++ b/assets/ai-optimizer/state.json @@ -0,0 +1,21 @@ +{ + "track": "gpu", + "current_model": "devstral-small-2:24b", + "model_index": 0, + "phase": "context_scaling", + "backend": "ollama", + "current_config": { + "num_ctx": 32768, + "num_gpu": 99, + "flash_attn": true + }, + "best_configs": { + "gpu": {}, + "ram": {} + }, + "completed_models": [], + "gpu_queue": ["devstral-small-2:24b", "qwen2.5-coder:32b", "codellama:34b-instruct"], + "ram_queue": ["qwen2.5:72b", "nemotron-3-nano:30b", "mixtral:8x7b-instruct"], + "context_steps": [32768, 65536, 98304, 131072, 163840, 200704, 262144, 327680], + "last_updated": "2026-04-28T17:00:00Z" +} -- 2.49.1 From 7ce0e46670e13778633776b8f938c93bb1d4a846 Mon Sep 17 00:00:00 2001 From: Hermes Agent Date: Tue, 28 Apr 2026 17:26:21 +0000 Subject: [PATCH 2/3] Consolidate cron job docs into single README.md --- assets/ai-optimizer/CRON_JOB_DRAFT.md | 283 ------------------ .../{CRON_EXECUTION_PROMPT.md => README.md} | 137 +++++---- 2 files changed, 74 insertions(+), 346 deletions(-) delete mode 100644 assets/ai-optimizer/CRON_JOB_DRAFT.md rename assets/ai-optimizer/{CRON_EXECUTION_PROMPT.md => README.md} (56%) diff --git a/assets/ai-optimizer/CRON_JOB_DRAFT.md b/assets/ai-optimizer/CRON_JOB_DRAFT.md deleted file mode 100644 index 3ff0781..0000000 --- a/assets/ai-optimizer/CRON_JOB_DRAFT.md +++ /dev/null @@ -1,283 +0,0 @@ -# AI Model Optimization Cron Job - -**Goal:** Find optimal configurations for maximum context size with full hardware utilization. - -**Hardware:** -- 2× AMD MI50 GPUs (32GB VRAM each, 64GB total) -- 128GB system RAM -- ROCm: HSA_OVERRIDE_GFX_VERSION=9.0.6, HIP_VISIBLE_DEVICES=0,1 - ---- - -## Model Queue - -### GPU-Optimized (Coding - prioritize speed + context on GPU) -1. `devstral-small-2:24b` - Best coding model -2. `qwen2.5-coder:32b` - Strong coder, fits on GPU+offload -3. `codellama:34b-instruct` - Legacy but solid - -### RAM-Optimized (Knowledge - prioritize max context, accept slower) -1. `qwen2.5:72b` - Best knowledge, needs heavy offload -2. `nemotron-3-nano:30b` - Good general knowledge -3. `mixtral:8x7b-instruct` - MoE, efficient for knowledge - ---- - -## Optimization Strategy - -**Two separate tracks:** - -### Track A: GPU-Focused (Coding) -``` -Baseline: num_ctx=32768, num_gpu=99, flash_attn=true -Steps: -1. Increase context: 32k → 65k → 98k → 131k → 163k -2. At each step, verify VRAM usage < 60GB (leave headroom) -3. If OOM: reduce num_gpu until stable, record best -4. Measure tokens/sec - if < 5 tok/s, consider context too high -``` - -### Track B: RAM-Focused (Knowledge) -``` -Baseline: num_ctx=65536, num_gpu=50, flash_attn=true -Steps: -1. Increase context: 65k → 131k → 200k → 262k → 327k -2. Allow heavy RAM offload (system RAM up to 100GB) -3. If OOM: reduce context or num_gpu -4. Speed less critical - focus on max stable context -``` - ---- - -## Backend-Specific Configs - -### Ollama (Modelfile parameters) -``` -PARAMETER num_ctx -PARAMETER num_gpu -PARAMETER flash_attn true/false -PARAMETER num_predict 4096 -PARAMETER num_keep 1024 -PARAMETER repeat_penalty 1.1 -``` - -### Llama.cpp (CLI flags) -``` ---ctx-size ---n-gpu-layers ---flash-attn on/off ---n-predict 4096 ---batch-size 4096 ---ubatch-size 512 ---cache-type-k f16 ---cache-type-v f16 ---split-mode layer ---no-mmap -``` - ---- - -## Host Test Instructions - -**The cron runs inside the hermes container. Some tests require host access:** - -### 1. VRAM Monitoring (HOST) -```bash -# Run on host to check VRAM usage during/after benchmark -sudo rocm-smi --showmeminfo vram - -# Or via docker exec if rocm-smi available in container -docker exec --privileged ollama rocm-smi --showmeminfo vram -``` - -### 2. Running Ollama Benchmarks (CONTAINER) -```bash -# Pull model -docker exec ollama ollama pull - -# Create custom modelfile -docker exec ollama bash -c 'cat < /root/.ollama/test.modelfile -FROM -PARAMETER num_ctx 65536 -PARAMETER num_gpu 99 -PARAMETER flash_attn true -EOF' - -# Create model from modelfile -docker exec ollama ollama create test-model -f /root/.ollama/test.modelfile - -# Run benchmark (warm model first) -docker exec ollama ollama run test-model "Write a Python async context manager with exponential backoff" - -# Cleanup -docker exec ollama ollama rm test-model -``` - -### 3. Running Llama.cpp Benchmarks (CONTAINER - needs llama.cpp container) -```bash -# Uncomment llama_cpp_devstral in compose.yml first -# Then rebuild: sudo nh os switch --flake .#lazyworkhorse - -# Test via HTTP API -curl http://localhost:8300/v1/completions \ - -H "Content-Type: application/json" \ - -d '{ - "model": "devstral-2-small-llama_cpp", - "prompt": "Write a Python function", - "max_tokens": 100 - }' -``` - -### 4. Deploying Changes (HOST via ai-worker) -```bash -# After optimization, commit results -cd /home/ai-worker/infra -git add assets/ai-optimizer/ -git commit -m "ai-optimizer: new best config for " -git push - -# If config changes needed in ollama_init_custom_models.nix: -# 1. Edit the file -# 2. nixpkgs-fmt . -# 3. Show diff to user -# 4. Wait for confirmation -# 5. sudo nh os switch --flake .#lazyworkhorse -``` - -### 5. Accessing Host from Hermes Container -```bash -# SSH to host as ai-worker (key should be mounted) -ssh -i /path/to/key ai-worker@host.docker.internal - -# Or via docker socket if mounted -# (not recommended for security) -``` - ---- - -## Benchmark Prompts - -### Coding (Track A) -``` -"Write a Python async context manager that retries a function with exponential backoff, max 5 retries, and logs each attempt using structlog. Include type hints and error handling." -``` - -### Knowledge (Track B) -``` -"Explain the complete memory hierarchy in modern GPUs, from registers through L1/L2 caches to VRAM, and how data moves between them during matrix multiplication. Include bandwidth considerations for each level." -``` - -### Measurement -- Tokens per second (generation speed) -- Time to first token (latency) -- VRAM usage (via rocm-smi) -- System RAM usage (via free -h) -- Context success (did it complete without OOM?) - ---- - -## State File Structure - -`/opt/data/infra/assets/ai-optimizer/state.json` - -```json -{ - "track": "gpu", - "current_model": "devstral-small-2:24b", - "model_index": 0, - "phase": "context_scaling", - "backend": "ollama", - "current_config": { - "num_ctx": 65536, - "num_gpu": 99, - "flash_attn": true - }, - "best_configs": { - "gpu": { - "devstral-small-2:24b": { - "backend": "ollama", - "num_ctx": 131072, - "num_gpu": 99, - "flash_attn": true, - "tokens_per_sec": 12.5, - "vram_used_gb": 58.2, - "tested_at": "2026-04-28T17:00:00Z" - } - }, - "ram": {} - }, - "completed_models": [], - "gpu_queue": ["devstral-small-2:24b", "qwen2.5-coder:32b", "codellama:34b-instruct"], - "ram_queue": ["qwen2.5:72b", "nemotron-3-nano:30b", "mixtral:8x7b-instruct"] -} -``` - ---- - -## Results CSV - -`/opt/data/infra/assets/ai-optimizer/results.csv` - -```csv -timestamp,track,model,backend,phase,num_ctx,num_gpu,flash_attn,tokens_per_sec,vram_gb,ram_gb,status,is_best -2026-04-28T17:00:00Z,gpu,devstral-small-2:24b,ollama,context_scaling,65536,99,true,15.2,52.1,18.4,success,false -``` - ---- - -## Cron Job Flow - -``` -1. Read state.json -2. If both queues empty → STOP (all models tested) -3. Select next model from current track queue -4. Pull model if needed (docker exec ollama ollama pull) -5. Create Modelfile / llama.cpp config with current test params -6. Run benchmark (both prompts) -7. Measure: tokens/sec, VRAM (rocm-smi), RAM (free -h) -8. If successful: - - Increase context (next step) - - Update current_config in state -9. If OOM/error: - - Record last good config as best_configs[track][model] - - Move to next model in queue -10. Update state.json -11. Append to results.csv -12. Git commit + push to /opt/data/infra -13. Send Matrix notification if available, else silent -``` - ---- - -## Matrix Notification (Optional) - -```python -# If matrix credentials available in environment -if os.getenv("MATRIX_HOME_SERVER") and os.getenv("MATRIX_ACCESS_TOKEN"): - # Send completion notification - # Room: !ai-optimizer:lazyworkhorse.net (or similar) - pass -# Else: silent, just commit -``` - ---- - -## Files to Create - -``` -/opt/data/infra/assets/ai-optimizer/ -├── state.json # Current progress -├── results.csv # All test results -├── best_configs.json # Final best configs (human-readable) -└── CRON_JOB_DRAFT.md # This file -``` - ---- - -## Notes - -- **No num_parallel**: Removed to avoid limiting other settings -- **Two tracks**: GPU (coding/speed) vs RAM (knowledge/context) -- **Both backends**: Test ollama first, then llama.cpp if available -- **Host tests**: rocm-smi must run on host or privileged container -- **Deploy**: ai-worker has sudo for nh/nixos-rebuild, must ask user first diff --git a/assets/ai-optimizer/CRON_EXECUTION_PROMPT.md b/assets/ai-optimizer/README.md similarity index 56% rename from assets/ai-optimizer/CRON_EXECUTION_PROMPT.md rename to assets/ai-optimizer/README.md index 20a6600..b7b2461 100644 --- a/assets/ai-optimizer/CRON_EXECUTION_PROMPT.md +++ b/assets/ai-optimizer/README.md @@ -1,12 +1,8 @@ -# AI Model Optimization Cron Job - EXECUTION PROMPT +# AI Model Optimization Cron Job -**When this cron runs, follow these instructions exactly:** +**Purpose:** Automatically find optimal ollama/llama.cpp configurations for maximum context size and hardware utilization. ---- - -## Your Role - -You are an AI model optimization agent. Your task is to find the best ollama/llama.cpp configuration for maximum context size and hardware utilization. +**Schedule:** Every hour **Hardware:** - 2× AMD MI50 GPUs (32GB VRAM each, 64GB total) @@ -18,9 +14,9 @@ You are an AI model optimization agent. Your task is to find the best ollama/lla ## File Locations ``` -STATE: /opt/data/infra/assets/ai-optimizer/state.json +STATE: /opt/data/infra/assets/ai-optimizer/state.json RESULTS: /opt/data/infra/assets/ai-optimizer/results.csv -INFRA_REPO: /opt/data/infra +REPO: /opt/data/infra (persistent - do not reclone) ``` --- @@ -46,6 +42,22 @@ INFRA_REPO: /opt/data/infra --- +## Optimization Strategy + +### GPU Track (Coding) +- Start: num_ctx=32768, num_gpu=99, flash_attn=true +- Increase context until OOM or tokens/sec < 5 +- Record best config before hitting wall +- Target: >10 tokens/sec with max context + +### RAM Track (Knowledge) +- Start: num_ctx=65536, num_gpu=50, flash_attn=true +- Allow heavy RAM offload (up to 100GB system RAM) +- Increase context until OOM +- Speed secondary to context size + +--- + ## Each Run - Step by Step ### 1. Read State @@ -56,9 +68,9 @@ cat assets/ai-optimizer/state.json ### 2. Determine Next Test - Read `track` (gpu or ram) -- Read `current_model` from queue at `model_index` -- Read `current_config` for parameters to test -- Select next context step from `context_steps` based on `phase` +- Get `current_model` from queue at `model_index` +- Get `current_config` for parameters to test +- Select next context step from `context_steps` ### 3. Pull Model (if needed) ```bash @@ -86,11 +98,10 @@ docker exec ollama ollama create test-model -f /root/.ollama/test_${model}.model docker exec ollama ollama run test-model "Hello" > /dev/null # Coding prompt -START=$(date +%s%N) docker exec ollama ollama run test-model "Write a Python async context manager that retries a function with exponential backoff, max 5 retries, and logs each attempt using structlog. Include type hints." -END=$(date +%s%N) -# Calculate tokens/sec from output +# Knowledge prompt +docker exec ollama ollama run test-model "Explain the complete memory hierarchy in modern GPUs, from registers through L1/L2 caches to VRAM, and how data moves between them during matrix multiplication." ``` ### 6. Measure VRAM (if possible) @@ -99,33 +110,25 @@ END=$(date +%s%N) rocm-smi --showmeminfo vram 2>/dev/null || \ # Try via docker docker exec --privileged ollama rocm-smi --showmeminfo vram 2>/dev/null || \ -# Fallback -echo "VRAM measurement unavailable" +echo "VRAM unavailable" ``` ### 7. Record Results - Parse tokens/sec from ollama output - Record VRAM/RAM usage -- Determine if this is best config so far for this model -- Update `best_configs` if tokens/sec improved or context increased +- Update `best_configs` if improved ### 8. Update State ```python -# Logic: if test_successful: if context_step < max_reached: - phase = "context_scaling" current_config.num_ctx = next_context_step else: - # Move to next model model_index += 1 - phase = "context_scaling" current_config.num_ctx = context_steps[0] else: - # OOM or error - record last good as best best_configs[track][current_model] = last_good_config model_index += 1 - phase = "context_scaling" ``` ### 9. Commit to Repo @@ -133,71 +136,79 @@ else: cd /opt/data/infra git add assets/ai-optimizer/ git commit -m "ai-optimizer: tested ${model} at ${num_ctx} ctx - ${status}" -git push origin master +git push ``` ### 10. Matrix Notification (if available) ```python import os if os.getenv("MATRIX_HOME_SERVER") and os.getenv("MATRIX_ACCESS_TOKEN"): - # Send notification to Matrix room - # Room ID from env or config + # Send notification pass # Else: silent ``` --- +## State File Structure + +```json +{ + "track": "gpu", + "current_model": "devstral-small-2:24b", + "model_index": 0, + "phase": "context_scaling", + "backend": "ollama", + "current_config": { + "num_ctx": 32768, + "num_gpu": 99, + "flash_attn": true + }, + "best_configs": { + "gpu": {}, + "ram": {} + }, + "completed_models": [], + "gpu_queue": ["devstral-small-2:24b", "qwen2.5-coder:32b", "codellama:34b-instruct"], + "ram_queue": ["qwen2.5:72b", "nemotron-3-nano:30b", "mixtral:8x7b-instruct"], + "context_steps": [32768, 65536, 98304, 131072, 163840, 200704, 262144, 327680], + "last_updated": "2026-04-28T17:00:00Z" +} +``` + +--- + +## Results CSV Format + +```csv +timestamp,track,model,backend,phase,num_ctx,num_gpu,flash_attn,tokens_per_sec,vram_gb,ram_gb,status,is_best +2026-04-28T17:00:00Z,gpu,devstral-small-2:24b,ollama,context_scaling,65536,99,true,15.2,52.1,18.4,success,false +``` + +--- + ## Stop Conditions 1. All models in both queues have `best_configs` recorded 2. Manual intervention needed (error in state.json `error` field) -3. No progress for 3 consecutive runs (stuck) +3. No progress for 3 consecutive runs --- ## Error Handling If any step fails: -1. Log error to state.json: `"error": {"message": "...", "timestamp": "..."}` +1. Log error: `"error": {"message": "...", "timestamp": "..."}` 2. Do NOT increment model_index (retry next run) 3. Commit state with error field 4. Exit gracefully --- -## Important Notes +## Notes -- **No num_parallel**: Do not use this parameter +- **No num_parallel**: Removed to avoid limiting other settings - **Two tracks**: Complete GPU track first, then RAM track -- **Backend**: Start with ollama, llama.cpp testing is optional (requires uncommenting in compose.yml) -- **Host access**: Some commands need host - use docker exec or SSH if available -- **Ask before deploy**: If config changes needed in NixOS modules, show diff and wait for user confirmation before `nh os switch` - ---- - -## Example State Transitions - -**Start:** -```json -{"track": "gpu", "model_index": 0, "current_model": "devstral-small-2:24b", "current_config": {"num_ctx": 32768, ...}} -``` - -**After successful test at 32k:** -```json -{"track": "gpu", "model_index": 0, "current_model": "devstral-small-2:24b", "current_config": {"num_ctx": 65536, ...}} -``` - -**After OOM at 131k:** -```json -{ - "track": "gpu", - "model_index": 1, - "current_model": "qwen2.5-coder:32b", - "best_configs": { - "gpu": { - "devstral-small-2:24b": {"num_ctx": 98304, "num_gpu": 99, "tokens_per_sec": 11.2} - } - } -} -``` +- **Backend**: Start with ollama, llama.cpp optional +- **Host access**: Use docker exec or SSH for rocm-smi +- **Ask before deploy**: Show diff before `nh os switch` -- 2.49.1 From 0ec198dec22f288e19db63fdd88be301e1756d05 Mon Sep 17 00:00:00 2001 From: Hermes Agent Date: Thu, 30 Apr 2026 16:07:05 +0000 Subject: [PATCH 3/3] feat: convert ai-optimizer from cron job to manual skill - Update README.md for manual execution workflow - Change model queue to deepseek-coder-v2:16b (better coding model) - Remove automated scheduling references - Add skill usage instructions for post-PR#1 merge --- assets/ai-optimizer/README.md | 135 ++++++++++++++++----------------- assets/ai-optimizer/state.json | 6 +- 2 files changed, 68 insertions(+), 73 deletions(-) diff --git a/assets/ai-optimizer/README.md b/assets/ai-optimizer/README.md index b7b2461..a0bf589 100644 --- a/assets/ai-optimizer/README.md +++ b/assets/ai-optimizer/README.md @@ -1,13 +1,13 @@ -# AI Model Optimization Cron Job +# AI Model Optimizer - Manual Skill -**Purpose:** Automatically find optimal ollama/llama.cpp configurations for maximum context size and hardware utilization. +**Purpose:** Find optimal ollama configurations for maximum context size and GPU utilization on AMD MI50 GPUs. -**Schedule:** Every hour +**Usage:** Run manually via Hermes skill when needed (not automated). **Hardware:** - 2× AMD MI50 GPUs (32GB VRAM each, 64GB total) - 128GB system RAM -- ROCm: HSA_OVERRIDE_GFX_VERSION=9.0.6, HIP_VISIBLE_DEVICES=0,1 +- ROCm: `HSA_OVERRIDE_GFX_VERSION=9.0.6`, `HIP_VISIBLE_DEVICES=0,1` --- @@ -21,21 +21,33 @@ REPO: /opt/data/infra (persistent - do not reclone) --- +## Quick Start + +```bash +# From Hermes container or any machine with ollama access +ollama-test-model --model devstral-small-2:24b --ctx 65536 +``` + +Or use the full workflow skill for systematic testing. + +--- + ## Model Queues ### GPU Track (Coding - prioritize speed + context on GPU) -1. `devstral-small-2:24b` -2. `qwen2.5-coder:32b` -3. `codellama:34b-instruct` +1. `deepseek-coder-v2:16b` - Best coding model, fits on GPU +2. `qwen2.5-coder:32b` - Alternative coding model +3. `codellama:34b-instruct` - Legacy option ### RAM Track (Knowledge - prioritize max context) -1. `qwen2.5:72b` -2. `nemotron-3-nano:30b` -3. `mixtral:8x7b-instruct` +1. `qwen2.5:72b` - Large knowledge model +2. `nemotron-3-nano:30b` - Efficient large model +3. `mixtral:8x7b-instruct` - MoE architecture --- ## Context Steps (in order) + ``` [32768, 65536, 98304, 131072, 163840, 200704, 262144, 327680] ``` @@ -45,45 +57,49 @@ REPO: /opt/data/infra (persistent - do not reclone) ## Optimization Strategy ### GPU Track (Coding) -- Start: num_ctx=32768, num_gpu=99, flash_attn=true +- Start: `num_ctx=32768`, `num_gpu=99`, `flash_attn=true` - Increase context until OOM or tokens/sec < 5 - Record best config before hitting wall - Target: >10 tokens/sec with max context ### RAM Track (Knowledge) -- Start: num_ctx=65536, num_gpu=50, flash_attn=true +- Start: `num_ctx=65536`, `num_gpu=50`, `flash_attn=true` - Allow heavy RAM offload (up to 100GB system RAM) - Increase context until OOM - Speed secondary to context size --- -## Each Run - Step by Step +## Manual Testing Workflow + +### 1. Quick Model Test + +```bash +# Test a model at specific context size +docker exec ollama ollama run : "Your prompt here" +``` + +### 2. Check Current State -### 1. Read State ```bash cd /opt/data/infra cat assets/ai-optimizer/state.json ``` -### 2. Determine Next Test -- Read `track` (gpu or ram) -- Get `current_model` from queue at `model_index` -- Get `current_config` for parameters to test -- Select next context step from `context_steps` - ### 3. Pull Model (if needed) + ```bash -docker exec ollama ollama list | grep -q "" || docker exec ollama ollama pull +docker exec ollama ollama pull : ``` ### 4. Create Test Modelfile + ```bash docker exec ollama bash -c "cat < /root/.ollama/test_${model}.modelfile FROM ${model} -PARAMETER num_ctx ${current_config.num_ctx} -PARAMETER num_gpu ${current_config.num_gpu} -PARAMETER flash_attn ${current_config.flash_attn} +PARAMETER num_ctx ${num_ctx} +PARAMETER num_gpu ${num_gpu} +PARAMETER flash_attn true PARAMETER num_predict 4096 PARAMETER num_keep 1024 PARAMETER repeat_penalty 1.1 @@ -93,6 +109,7 @@ docker exec ollama ollama create test-model -f /root/.ollama/test_${model}.model ``` ### 5. Run Benchmark + ```bash # Warm up docker exec ollama ollama run test-model "Hello" > /dev/null @@ -104,7 +121,8 @@ docker exec ollama ollama run test-model "Write a Python async context manager t docker exec ollama ollama run test-model "Explain the complete memory hierarchy in modern GPUs, from registers through L1/L2 caches to VRAM, and how data moves between them during matrix multiplication." ``` -### 6. Measure VRAM (if possible) +### 6. Measure VRAM + ```bash # Try host first rocm-smi --showmeminfo vram 2>/dev/null || \ @@ -114,24 +132,14 @@ echo "VRAM unavailable" ``` ### 7. Record Results -- Parse tokens/sec from ollama output -- Record VRAM/RAM usage -- Update `best_configs` if improved -### 8. Update State -```python -if test_successful: - if context_step < max_reached: - current_config.num_ctx = next_context_step - else: - model_index += 1 - current_config.num_ctx = context_steps[0] -else: - best_configs[track][current_model] = last_good_config - model_index += 1 -``` +Update `state.json` and append to `results.csv`: +- tokens/sec from ollama output +- VRAM/RAM usage +- Whether this config is the new best + +### 8. Commit Changes -### 9. Commit to Repo ```bash cd /opt/data/infra git add assets/ai-optimizer/ @@ -139,15 +147,6 @@ git commit -m "ai-optimizer: tested ${model} at ${num_ctx} ctx - ${status}" git push ``` -### 10. Matrix Notification (if available) -```python -import os -if os.getenv("MATRIX_HOME_SERVER") and os.getenv("MATRIX_ACCESS_TOKEN"): - # Send notification - pass -# Else: silent -``` - --- ## State File Structure @@ -155,7 +154,7 @@ if os.getenv("MATRIX_HOME_SERVER") and os.getenv("MATRIX_ACCESS_TOKEN"): ```json { "track": "gpu", - "current_model": "devstral-small-2:24b", + "current_model": "deepseek-coder-v2:16b", "model_index": 0, "phase": "context_scaling", "backend": "ollama", @@ -169,10 +168,10 @@ if os.getenv("MATRIX_HOME_SERVER") and os.getenv("MATRIX_ACCESS_TOKEN"): "ram": {} }, "completed_models": [], - "gpu_queue": ["devstral-small-2:24b", "qwen2.5-coder:32b", "codellama:34b-instruct"], + "gpu_queue": ["deepseek-coder-v2:16b", "qwen2.5-coder:32b", "codellama:34b-instruct"], "ram_queue": ["qwen2.5:72b", "nemotron-3-nano:30b", "mixtral:8x7b-instruct"], "context_steps": [32768, 65536, 98304, 131072, 163840, 200704, 262144, 327680], - "last_updated": "2026-04-28T17:00:00Z" + "last_updated": "2026-04-30T00:00:00Z" } ``` @@ -182,33 +181,29 @@ if os.getenv("MATRIX_HOME_SERVER") and os.getenv("MATRIX_ACCESS_TOKEN"): ```csv timestamp,track,model,backend,phase,num_ctx,num_gpu,flash_attn,tokens_per_sec,vram_gb,ram_gb,status,is_best -2026-04-28T17:00:00Z,gpu,devstral-small-2:24b,ollama,context_scaling,65536,99,true,15.2,52.1,18.4,success,false +2026-04-30T00:00:00Z,gpu,deepseek-coder-v2:16b,ollama,context_scaling,65536,99,true,15.2,52.1,18.4,success,false ``` --- -## Stop Conditions +## Skill Usage -1. All models in both queues have `best_configs` recorded -2. Manual intervention needed (error in state.json `error` field) -3. No progress for 3 consecutive runs +Once PR #1 (ai-worker-restricted-access) is merged: ---- +```bash +# From Hermes container, SSH to host for direct ollama access +ssh -i /path/to/key ai-worker@host docker exec ollama ollama run -## Error Handling - -If any step fails: -1. Log error: `"error": {"message": "...", "timestamp": "..."}` -2. Do NOT increment model_index (retry next run) -3. Commit state with error field -4. Exit gracefully +# Or run the skill directly +ollama-benchmark --model deepseek-coder-v2:16b --track gpu +``` --- ## Notes -- **No num_parallel**: Removed to avoid limiting other settings -- **Two tracks**: Complete GPU track first, then RAM track -- **Backend**: Start with ollama, llama.cpp optional +- **Manual execution only** - No cron job, run when needed +- **Two tracks**: Complete GPU track first (coding models), then RAM track +- **Backend**: ollama (llama.cpp optional for advanced users) - **Host access**: Use docker exec or SSH for rocm-smi -- **Ask before deploy**: Show diff before `nh os switch` +- **Commit results**: Push best configs to repo for reference diff --git a/assets/ai-optimizer/state.json b/assets/ai-optimizer/state.json index fff69f9..41b6d04 100644 --- a/assets/ai-optimizer/state.json +++ b/assets/ai-optimizer/state.json @@ -1,6 +1,6 @@ { "track": "gpu", - "current_model": "devstral-small-2:24b", + "current_model": "deepseek-coder-v2:16b", "model_index": 0, "phase": "context_scaling", "backend": "ollama", @@ -14,8 +14,8 @@ "ram": {} }, "completed_models": [], - "gpu_queue": ["devstral-small-2:24b", "qwen2.5-coder:32b", "codellama:34b-instruct"], + "gpu_queue": ["deepseek-coder-v2:16b", "qwen2.5-coder:32b", "codellama:34b-instruct"], "ram_queue": ["qwen2.5:72b", "nemotron-3-nano:30b", "mixtral:8x7b-instruct"], "context_steps": [32768, 65536, 98304, 131072, 163840, 200704, 262144, 327680], - "last_updated": "2026-04-28T17:00:00Z" + "last_updated": "2026-04-30T00:00:00Z" } -- 2.49.1