refactor(benchmark): remove Handoff documentation and update benchmark metrics

- Deleted the HANDOFF.md file as it is no longer needed.
- Updated model selection logic in 03_benchmark.yml to ensure accurate metrics display.
- Enhanced evaluation metrics to handle cases where model selections may not exist, preventing potential errors in output.

HANDOFF.md

@@ -1,60 +0,0 @@
-# Session Handoff — 2026-03-09
-
-## Branch
-
-`feature/three-pass-benchmark`
-
-## What Was Just Done
-
-Fixed two bugs that caused near-zero tok/sec and latency=9999 in the concurrent benchmark.
-
-### Bug 1 — Queue contamination (tok/sec ≈ 0.04–0.08)
-
-**Root cause:** Benchmark was firing all 21 requests (3 models × 7 prompts) at once via
-async uri. With `OLLAMA_NUM_PARALLEL=2`, only 2 slots run at a time; the other 19 queue.
-`eval_duration` includes queue-wait time → tok/sec measured as ~0.18 instead of ~22.
-
-**Fix:** `playbooks/_bench_tier_batch.yml` — replaced the 7 async fire+collect tasks
-with 5 synchronous `uri` tasks (no `async`/`poll`). Node1 runs all its prompts first,
-then node0. One request at a time = idle slot = clean eval_duration.
-
-Key structural changes:
-
-- No more `_bench_node1_jobs` / `_bench_node0_jobs` intermediate registers
-- No more `async_status` collect tasks
-- Accumulate tasks now use `item.item` (top-level in sync uri) instead of
-`item._async_job.item` (the old async indirection)
-- `timeout` reverted from `benchmark_large_timeout × 15` to plain `benchmark_large_timeout`
-
-### Bug 2 — latency_ms = 9999 (latency_ns = 0)
-
-**Root cause:** Load phase warm-up "Hi" populates KV cache. Benchmark "Hi" hits cache →
-`prompt_eval_duration ≈ 0` and `eval_duration ≈ 0` → sum = 0 → latency_ms = 9999.
-
-**Fix:** `playbooks/03_benchmark.yml` line 268 — changed latency measurement from
-`eval_duration + prompt_eval_duration` to `resp.total_duration | default(0) | int`.
-`total_duration` is true wall-clock end-to-end; never zero for a completed request.
-
-## State of the Branch
-
-Both fixes are committed as `d9a991f`. Nothing is pending.
-
-## Expected Results After Fix
-
-- `avg_tok_per_sec` ≈ 5–25 (was 0.04–0.08)
-- `latency_ms` ≈ 300–6000 ms (was 9999)
-- Composite scores ≈ 0.50–0.85 (was ≈ 0.45 flat)
-
-## Next Steps
-
-1. Run benchmark to verify the fix:
-
-1. If scores look correct, update warm-up slots:
-
-```bash
-ansible-playbook playbooks/03_benchmark.yml -K -e @local.yml && \
-ansible-playbook playbooks/04_models.yml -K -e @local.yml
-```
-
-1. Merge `feature/three-pass-benchmark` into `master` when satisfied.
-

playbooks/03_benchmark.yml

@@ -393,12 +393,12 @@
           ## Model Selection (6-slot / 2-socket)
           | Slot | Socket | Role | Model | Composite Score |
           |------|--------|------|-------|----------------|
-          | 1 | Node 1 (port 11434) | General (locked) | {{ selection.slot1_general }} | {{ parsed_metrics[selection.slot1_general].general_composite | default('N/A') }} |
-          | 2 | Node 1 (port 11434) | General (locked) | {{ selection.slot2_general }} | {{ parsed_metrics[selection.slot2_general].general_composite | default('N/A') }} |
-          | 5 | Node 1 (port 11434) | General (rotate) | {{ selection.slot5_general_rotate }} | {{ parsed_metrics[selection.slot5_general_rotate].general_composite | default('N/A') }} |
-          | 3 | Node 0 (port 11435) | Coding (locked) | {{ selection.slot3_coding }} | {{ parsed_metrics[selection.slot3_coding].coding_composite | default('N/A') }} |
-          | 4 | Node 0 (port 11435) | Coding (locked) | {{ selection.slot4_coding }} | {{ parsed_metrics[selection.slot4_coding].coding_composite | default('N/A') }} |
-          | 6 | Node 0 (port 11435) | Coding (rotate) | {{ selection.slot6_coding_rotate }} | {{ parsed_metrics[selection.slot6_coding_rotate].coding_composite | default('N/A') }} |
+          | 1 | Node 1 (port 11434) | General (locked) | {{ selection.slot1_general }} | {{ (parsed_metrics[selection.slot1_general].general_composite if selection.slot1_general in parsed_metrics else 'N/A') }} |
+          | 2 | Node 1 (port 11434) | General (locked) | {{ selection.slot2_general }} | {{ (parsed_metrics[selection.slot2_general].general_composite if selection.slot2_general in parsed_metrics else 'N/A') }} |
+          | 5 | Node 1 (port 11434) | General (rotate) | {{ selection.slot5_general_rotate }} | {{ (parsed_metrics[selection.slot5_general_rotate].general_composite if selection.slot5_general_rotate in parsed_metrics else 'N/A') }} |
+          | 3 | Node 0 (port 11435) | Coding (locked) | {{ selection.slot3_coding }} | {{ (parsed_metrics[selection.slot3_coding].coding_composite if selection.slot3_coding in parsed_metrics else 'N/A') }} |
+          | 4 | Node 0 (port 11435) | Coding (locked) | {{ selection.slot4_coding }} | {{ (parsed_metrics[selection.slot4_coding].coding_composite if selection.slot4_coding in parsed_metrics else 'N/A') }} |
+          | 6 | Node 0 (port 11435) | Coding (rotate) | {{ selection.slot6_coding_rotate }} | {{ (parsed_metrics[selection.slot6_coding_rotate].coding_composite if selection.slot6_coding_rotate in parsed_metrics else 'N/A') }} |
 
           ## Detailed Metrics
           {% for model, metrics in parsed_metrics.items() %}

templates/ollama/ollama-node0.service.j2

@@ -7,7 +7,7 @@ Wants=network-online.target
 ExecStart=/usr/bin/numactl --cpunodebind=0 {{ ollama_binary_path }} serve
 Environment="OLLAMA_API_KEY={{ ollama_api_key }}"
 Environment="OLLAMA_HOST=0.0.0.0:{{ ollama_node0_port }}"
-Environment="OLLAMA_MODELS={{ ollama_models_path }}"
+Environment="OLLAMA_MODELS=/mnt/ai_data/ollama_models"
 Environment="OLLAMA_KEEP_ALIVE={{ ollama_keep_alive }}"
 Environment="OLLAMA_FLASH_ATTENTION={{ ollama_flash_attention }}"
 Environment="OLLAMA_NUM_THREADS={{ ollama_num_threads }}"