Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76621.60222424015,
-      "required_pages_per_second_upper_estimate": 0.005300860228484309
+      "required_token_positions_per_second": 76126.38950607565,
+      "required_pages_per_second_upper_estimate": 0.005266600263589874
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9577.700278030019,
-      "required_pages_per_second_upper_estimate": 0.004009983101286023
+      "required_token_positions_per_second": 9515.798688259456,
+      "required_pages_per_second_upper_estimate": 0.003984066198301274
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19155.400556060038,
-      "required_pages_per_second_upper_estimate": 0.003989564773943978
+      "required_token_positions_per_second": 19031.597376518912,
+      "required_pages_per_second_upper_estimate": 0.0039637798365549605
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28733.100834090055,
-      "required_pages_per_second_upper_estimate": 0.0017225055220990487
+      "required_token_positions_per_second": 28547.396064778368,
+      "required_pages_per_second_upper_estimate": 0.0017113728047335774
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28733.100834090055,
-      "required_pages_per_second_upper_estimate": 0.0022338722062965584
+      "required_token_positions_per_second": 28547.396064778368,
+      "required_pages_per_second_upper_estimate": 0.00221943447731153
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28733.100834090055,
-      "required_pages_per_second_upper_estimate": 0.368557366299689
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28547.396064778368,
+      "required_pages_per_second_upper_estimate": 0.3661753449132049
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08208887971852619,
-    "producer_p95_seconds": 0.2956290830006765,
-    "gpu_update_seconds": 0.962235164232513,
-    "producer_batches_per_second": 12.181917982422112,
-    "gpu_batches_per_second": 1.0392469919737433,
-    "producer_headroom": 11.721869850482944,
+    "producer_batch_seconds": 0.40220647534380305,
+    "producer_p95_seconds": 0.30714974800139316,
+    "gpu_update_seconds": 0.9684946373834762,
+    "producer_batches_per_second": 2.4862851826172307,
+    "gpu_batches_per_second": 1.0325302396114862,
+    "producer_headroom": 2.4079538663707845,
     "sustainable": true,
-    "buffer_seconds": 60.62081534664832,
+    "buffer_seconds": 60.046667517775525,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 332490.01122666814,
+    "generation_bytes_per_second": 306918.25497376,
     "upload_bytes_per_second": 6060519.67151889,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.34461643791231,
+    "estimated_full_shards_per_hour": 65.85751282605743,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -258,3 +258,4 @@
 {"step": 510800, "loss": 0.5592138767242432, "lr": 0.0003, "grad_norm": 0.15223429740935868, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74763728380, "tokens_per_second": 150930.57722491908, "data_wait_s": 1.7693000700091943e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 510900, "loss": 0.5510307550430298, "lr": 0.0003, "grad_norm": 0.16456415781890485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74778295326, "tokens_per_second": 151104.70404062924, "data_wait_s": 1.678300031926483e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 511000, "loss": 0.5631042718887329, "lr": 0.0003, "grad_norm": 0.1702798864645056, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74793099178, "tokens_per_second": 153885.9698552784, "data_wait_s": 1.3248998584458604e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 511100, "loss": 0.5531523823738098, "lr": 0.0003, "grad_norm": 0.15254929866275538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74807754497, "tokens_per_second": 152788.15052105548, "data_wait_s": 1.9394999981159344e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:834905fba9cfc6a020812a73a40b4dc788828472500a4f42a9de18df09e00320
-size 554760
+oid sha256:65f435eaa4d69bb3d7f0db07eb19aaae9ce7157381019709ac941b516aa177c0
+size 556878