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": 77309.70176116753,
-      "required_pages_per_second_upper_estimate": 0.00534846455105984
+      "required_token_positions_per_second": 76270.45609451138,
+      "required_pages_per_second_upper_estimate": 0.005276567124458398
     },
     {
       "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": 9663.71272014594,
-      "required_pages_per_second_upper_estimate": 0.004045994714655921
+      "required_token_positions_per_second": 9533.807011813922,
+      "required_pages_per_second_upper_estimate": 0.003991605907316971
     },
     {
       "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": 19327.42544029188,
-      "required_pages_per_second_upper_estimate": 0.0040253930207282
+      "required_token_positions_per_second": 19067.614023627844,
+      "required_pages_per_second_upper_estimate": 0.003971281154324894
     },
     {
       "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": 28991.13816043782,
-      "required_pages_per_second_upper_estimate": 0.0017379744658134132
+      "required_token_positions_per_second": 28601.421035441766,
+      "required_pages_per_second_upper_estimate": 0.0017146115192334943
     },
     {
       "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": 28991.13816043782,
-      "required_pages_per_second_upper_estimate": 0.0022539334734338473
+      "required_token_positions_per_second": 28601.421035441766,
+      "required_pages_per_second_upper_estimate": 0.0022236346811498765
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28991.13816043782,
-      "required_pages_per_second_upper_estimate": 0.3718671920631831
+      "required_token_positions_per_second": 28601.421035441766,
+      "required_pages_per_second_upper_estimate": 0.3668683192293809
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11502308989020094,
-    "producer_p95_seconds": 0.3546597379972809,
-    "gpu_update_seconds": 0.953670733690935,
-    "producer_batches_per_second": 8.693906596967468,
-    "gpu_batches_per_second": 1.0485799392519466,
-    "producer_headroom": 8.291124282970426,
+    "producer_batch_seconds": 0.4803758457497338,
+    "producer_p95_seconds": 0.33194501799880527,
+    "gpu_update_seconds": 0.9666652564479113,
+    "producer_batches_per_second": 2.0817033346863574,
+    "gpu_batches_per_second": 1.0344842677749482,
+    "producer_headroom": 2.01231028787306,
     "sustainable": true,
-    "buffer_seconds": 60.08125622252891,
+    "buffer_seconds": 60.899911156218415,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 302628.9121667575,
-    "upload_bytes_per_second": 5013223.303476246,
+    "generation_bytes_per_second": 253537.88037470382,
+    "upload_bytes_per_second": 5016862.906902537,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.9371197104649,
+    "estimated_full_shards_per_hour": 54.40332706862293,
     "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": 1.2127139630028978,
-      "bytes": 7187251,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.2223667069993098,
       "bytes": 5868379,
@@ -562,6 +556,12 @@
       "bytes": 7496236,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.0467049810031313,
+      "bytes": 6666834,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -372,3 +372,4 @@
 {"step": 594300, "loss": 0.5571348071098328, "lr": 0.0003, "grad_norm": 0.20933298242937576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86987617518, "tokens_per_second": 153495.50530834612, "data_wait_s": 1.3780001609120518e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 594400, "loss": 0.5700617432594299, "lr": 0.0003, "grad_norm": 0.15909297899160507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87002080578, "tokens_per_second": 150434.29706185515, "data_wait_s": 1.0928000847343355e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 594500, "loss": 0.5499181747436523, "lr": 0.0003, "grad_norm": 0.14162799576614568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87016766650, "tokens_per_second": 152427.0866842106, "data_wait_s": 1.3689001207239926e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 594600, "loss": 0.6324679255485535, "lr": 0.0003, "grad_norm": 0.23020920522455954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87031483910, "tokens_per_second": 150174.36969955018, "data_wait_s": 1.0929994459729642e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8a0c3f14d20d660369436cfde9c736e5eb4ad352a0f7c1528786fefe96297315
-size 797778
+oid sha256:4fa9ad988a55a79957091c75d5763ac5b8aa24daf4d88a5925a2fde6ad1f8542
+size 799896