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": 76680.47853736079,
-      "required_pages_per_second_upper_estimate": 0.005304933428437862
+      "required_token_positions_per_second": 75832.22366015913,
+      "required_pages_per_second_upper_estimate": 0.005246249187810568
     },
     {
       "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": 9585.059817170099,
-      "required_pages_per_second_upper_estimate": 0.004013064386639321
+      "required_token_positions_per_second": 9479.027957519891,
+      "required_pages_per_second_upper_estimate": 0.003968671061200793
     },
     {
       "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": 19170.119634340197,
-      "required_pages_per_second_upper_estimate": 0.003992630369781537
+      "required_token_positions_per_second": 18958.055915039782,
+      "required_pages_per_second_upper_estimate": 0.003948463089547618
     },
     {
       "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": 28755.179451510296,
-      "required_pages_per_second_upper_estimate": 0.0017238291015012953
+      "required_token_positions_per_second": 28437.083872559673,
+      "required_pages_per_second_upper_estimate": 0.0017047597572470315
     },
     {
       "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": 28755.179451510296,
-      "required_pages_per_second_upper_estimate": 0.0022355887216874075
+      "required_token_positions_per_second": 28437.083872559673,
+      "required_pages_per_second_upper_estimate": 0.002210858189578581
     },
     {
       "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": 28755.179451510296,
-      "required_pages_per_second_upper_estimate": 0.36884056709778346
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28437.083872559673,
+      "required_pages_per_second_upper_estimate": 0.36476037855542737
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.12698193352341036,
-    "producer_p95_seconds": 0.31068587499976275,
-    "gpu_update_seconds": 0.9614963470015089,
-    "producer_batches_per_second": 7.875136031186989,
-    "gpu_batches_per_second": 1.0400455530783528,
-    "producer_headroom": 7.571914526126251,
+    "producer_batch_seconds": 0.08594841671902032,
+    "producer_p95_seconds": 0.3017171040009998,
+    "gpu_update_seconds": 0.9722515896462541,
+    "producer_batches_per_second": 11.63488564622623,
+    "gpu_batches_per_second": 1.0285403599739467,
+    "producer_headroom": 11.312036064895837,
     "sustainable": true,
-    "buffer_seconds": 60.57426986109506,
+    "buffer_seconds": 60.279598558067754,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329072.0058392974,
-    "upload_bytes_per_second": 6182369.709201382,
+    "generation_bytes_per_second": 339437.1945102421,
+    "upload_bytes_per_second": 6151997.828176973,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.61119204887572,
+    "estimated_full_shards_per_hour": 72.83532024841735,
     "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": [
-    {
-      "seconds": 1.181685737999942,
-      "bytes": 6229682,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.7940784490019723,
       "bytes": 968621,
@@ -562,6 +556,12 @@
       "bytes": 12466956,
       "pages": 47,
       "shards": 3
+    },
+    {
+      "seconds": 0.9538209790007386,
+      "bytes": 2685432,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -247,3 +247,4 @@
 {"step": 509700, "loss": 0.6125875115394592, "lr": 0.0003, "grad_norm": 0.17960290333110626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74602682646, "tokens_per_second": 141249.430933694, "data_wait_s": 1.763299951562658e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 509800, "loss": 0.5527863502502441, "lr": 0.0003, "grad_norm": 0.1797644494273301, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74617502095, "tokens_per_second": 153857.28984165288, "data_wait_s": 2.068699905066751e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 509900, "loss": 0.6094302535057068, "lr": 0.0003, "grad_norm": 0.17782631633133775, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74632221388, "tokens_per_second": 152602.52908936288, "data_wait_s": 1.872799839475192e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 510000, "loss": 0.5670058131217957, "lr": 0.0003, "grad_norm": 0.16512437787146803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74646740779, "tokens_per_second": 151310.01243005274, "data_wait_s": 1.5210000128718093e-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 sha256:7cc4569ace9e373fb5b1116bd670cb8435d0999ca7b31dd1c6e77444507818bf
-size 530679
+oid sha256:08d50e0968487d91925c987d981dd19c84f1c1b394b0830e4bbb39e86254fe13
+size 532797