Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76287.98759051682,
-      "required_pages_per_second_upper_estimate": 0.006612449414209142
+      "required_token_positions_per_second": 75792.03665743186,
+      "required_pages_per_second_upper_estimate": 0.006569461644305486
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9535.998448814602,
-      "required_pages_per_second_upper_estimate": 0.0038040523571144895
+      "required_token_positions_per_second": 9474.004582178983,
+      "required_pages_per_second_upper_estimate": 0.0037793220768226354
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19071.996897629204,
-      "required_pages_per_second_upper_estimate": 0.004411970826585183
+      "required_token_positions_per_second": 18948.009164357965,
+      "required_pages_per_second_upper_estimate": 0.004383288446602459
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28607.995346443804,
-      "required_pages_per_second_upper_estimate": 0.001742076822046763
+      "required_token_positions_per_second": 28422.01374653695,
+      "required_pages_per_second_upper_estimate": 0.0017307515183824789
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28607.995346443804,
-      "required_pages_per_second_upper_estimate": 0.0022023226737688767
+      "required_token_positions_per_second": 28422.01374653695,
+      "required_pages_per_second_upper_estimate": 0.002188005295378028
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28607.995346443804,
-      "required_pages_per_second_upper_estimate": 0.35521555740149
+      "required_token_positions_per_second": 28422.01374653695,
+      "required_pages_per_second_upper_estimate": 0.3529062883710214
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.060224770640630254,
-    "producer_p95_seconds": 0.20124438499988173,
-    "gpu_update_seconds": 0.9664431102278147,
-    "producer_batches_per_second": 16.604463402063278,
-    "gpu_batches_per_second": 1.0347220539078343,
-    "producer_headroom": 16.047269253953957,
+    "producer_batch_seconds": 0.07963236264845364,
+    "producer_p95_seconds": 0.21319815900096728,
+    "gpu_update_seconds": 0.9727671039272768,
+    "producer_batches_per_second": 12.557708533835882,
+    "gpu_batches_per_second": 1.027995288864907,
+    "producer_headroom": 12.21572576242238,
     "sustainable": true,
-    "buffer_seconds": 60.88591594435233,
+    "buffer_seconds": 60.31156044349116,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373726.7518663968,
-    "upload_bytes_per_second": 5566098.0187622635,
+    "generation_bytes_per_second": 327051.73034477443,
+    "upload_bytes_per_second": 5504549.488057209,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.19306103700569,
+    "estimated_full_shards_per_hour": 70.17768795735765,
     "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": [
@@ -508,6 +508,12 @@
       "bytes": 7450954,
       "pages": 28,
       "shards": 3
+    },
+    {
+      "seconds": 0.8681581979999464,
+      "bytes": 818733,
+      "pages": 23,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -66,3 +66,4 @@
 {"step": 286800, "loss": 0.5679254066199064, "lr": 0.0003, "grad_norm": 0.15646814916094784, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41979241507, "tokens_per_second": 147997.23661188598, "data_wait_s": 1.4410999938263558e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 286900, "loss": 0.5590508412569761, "lr": 0.0003, "grad_norm": 0.14929049746275003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41993814595, "tokens_per_second": 149200.73622699492, "data_wait_s": 1.3671000488102436e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 287000, "loss": 0.5955741610378027, "lr": 0.0003, "grad_norm": 0.15505631039645046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42008581843, "tokens_per_second": 150827.1468653094, "data_wait_s": 1.0313000530004501e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 287100, "loss": 0.5854803398251534, "lr": 0.0003, "grad_norm": 0.14064032562833925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42023205706, "tokens_per_second": 149778.58866883282, "data_wait_s": 1.8060000002151355e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9a653ece26cf1a5c1b8edc56b1dfe80002365a6c9880b95556946ec7768233d5
-size 144972
+oid sha256:a29ff92f98aa38bd9a50ad0364ade6b02437776716176bf0810ba6027dc266ae
+size 147090