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": 76379.6502437578,
-      "required_pages_per_second_upper_estimate": 0.0052841214290685985
+      "required_token_positions_per_second": 76098.72172319028,
+      "required_pages_per_second_upper_estimate": 0.005264686142171762
     },
     {
       "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": 9547.456280469725,
-      "required_pages_per_second_upper_estimate": 0.003997320571074015
+      "required_token_positions_per_second": 9512.340215398784,
+      "required_pages_per_second_upper_estimate": 0.003982618207935638
     },
     {
       "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": 19094.91256093945,
-      "required_pages_per_second_upper_estimate": 0.003976966719736061
+      "required_token_positions_per_second": 19024.68043079757,
+      "required_pages_per_second_upper_estimate": 0.00396233921917333
     },
     {
       "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": 28642.368841409178,
-      "required_pages_per_second_upper_estimate": 0.0017170662776775612
+      "required_token_positions_per_second": 28537.020646196353,
+      "required_pages_per_second_upper_estimate": 0.0017107508142319326
     },
     {
       "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": 28642.368841409178,
-      "required_pages_per_second_upper_estimate": 0.002226818192953479
+      "required_token_positions_per_second": 28537.020646196353,
+      "required_pages_per_second_upper_estimate": 0.0022186278341534216
     },
     {
       "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": 28642.368841409178,
-      "required_pages_per_second_upper_estimate": 0.3673935537180023
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28537.020646196353,
+      "required_pages_per_second_upper_estimate": 0.36604226018389135
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.1218357356014792,
-    "producer_p95_seconds": 0.2170104420001735,
-    "gpu_update_seconds": 0.9652832890004689,
-    "producer_batches_per_second": 8.207772498464392,
-    "gpu_batches_per_second": 1.0359653082106908,
-    "producer_headroom": 7.922825632685305,
+    "producer_batch_seconds": 0.09755961284376724,
+    "producer_p95_seconds": 0.21153871999922558,
+    "gpu_update_seconds": 0.9688467602410747,
+    "producer_batches_per_second": 10.25014317760166,
+    "gpu_batches_per_second": 1.0321549712889306,
+    "producer_headroom": 9.930818009626522,
     "sustainable": true,
-    "buffer_seconds": 60.81284720702954,
+    "buffer_seconds": 60.068499134946634,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344117.2349695751,
-    "upload_bytes_per_second": 3888073.6270537637,
+    "generation_bytes_per_second": 343858.3865841944,
+    "upload_bytes_per_second": 3964553.9177787444,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.83954798522414,
+    "estimated_full_shards_per_hour": 73.78400514740346,
     "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": 1.6723634379995929,
-      "bytes": 50251,
-      "pages": 1,
-      "shards": 1
-    },
     {
       "seconds": 2.1410375759996896,
       "bytes": 10006036,
@@ -562,6 +556,12 @@
       "bytes": 1164036,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 2.5616059220010357,
+      "bytes": 11807815,
+      "pages": 42,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -118,3 +118,4 @@
 {"step": 811800, "loss": 0.5481447577476501, "lr": 3e-05, "grad_norm": 0.2835895224033298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118824178708, "tokens_per_second": 152101.93591095626, "data_wait_s": 1.542999962111935e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 811900, "loss": 0.5262207984924316, "lr": 3e-05, "grad_norm": 0.2765897276631927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118838740313, "tokens_per_second": 150479.90507845033, "data_wait_s": 1.8888000340666622e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 812000, "loss": 0.5742809176445007, "lr": 3e-05, "grad_norm": 0.3177930254943342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118853246833, "tokens_per_second": 149470.66895386908, "data_wait_s": 1.7865000700112432e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 812100, "loss": 0.535494327545166, "lr": 3e-05, "grad_norm": 0.27763589498637725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118867908223, "tokens_per_second": 151764.48540176268, "data_wait_s": 2.0464998669922352e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c80fc9f7417bbc8c00048308103f4262893d1032163b00511b60d7f164ce6d48
-size 255891
+oid sha256:b81100c5dcab35b8ad6cad4c6d7e8a04c7b8b6b8c44f5f19f93e61241a0c8b27
+size 258009