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": 77196.25219599612,
-      "required_pages_per_second_upper_estimate": 0.005340615846902023
+      "required_token_positions_per_second": 77281.68919243415,
+      "required_pages_per_second_upper_estimate": 0.005346526576556751
     },
     {
       "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": 9649.531524499514,
-      "required_pages_per_second_upper_estimate": 0.004040057344175803
+      "required_token_positions_per_second": 9660.211149054268,
+      "required_pages_per_second_upper_estimate": 0.004044528679960958
     },
     {
       "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": 19299.06304899903,
-      "required_pages_per_second_upper_estimate": 0.004019485882588456
+      "required_token_positions_per_second": 19320.422298108537,
+      "required_pages_per_second_upper_estimate": 0.004023934450896688
     },
     {
       "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": 28948.59457349854,
-      "required_pages_per_second_upper_estimate": 0.0017354240427366993
+      "required_token_positions_per_second": 28980.633447162807,
+      "required_pages_per_second_upper_estimate": 0.0017373447242923447
     },
     {
       "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": 28948.59457349854,
-      "required_pages_per_second_upper_estimate": 0.002250625896678781
+      "required_token_positions_per_second": 28980.633447162807,
+      "required_pages_per_second_upper_estimate": 0.002253116778182066
     },
     {
       "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": 28948.59457349854,
-      "required_pages_per_second_upper_estimate": 0.37132148860967074
+      "required_token_positions_per_second": 28980.633447162807,
+      "required_pages_per_second_upper_estimate": 0.3717324488803736
     }
   ],
   "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": 0.1234200835078525,
-    "producer_p95_seconds": 0.2275281330003054,
-    "gpu_update_seconds": 0.9550722723275419,
-    "producer_batches_per_second": 8.10240903731341,
-    "gpu_batches_per_second": 1.047041181043784,
-    "producer_headroom": 7.738386210594131,
+    "producer_batch_seconds": 0.06195511435953449,
+    "producer_p95_seconds": 0.2147299850003037,
+    "gpu_update_seconds": 0.9540164141134992,
+    "producer_batches_per_second": 16.14071752328396,
+    "gpu_batches_per_second": 1.0481999944720344,
+    "producer_headroom": 15.398509452782283,
     "sustainable": true,
-    "buffer_seconds": 60.16955315663514,
+    "buffer_seconds": 60.103034089150455,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 395171.39919200586,
-    "upload_bytes_per_second": 4647958.401071017,
+    "generation_bytes_per_second": 379993.02251748345,
+    "upload_bytes_per_second": 4693839.307089437,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.79458314724094,
+    "estimated_full_shards_per_hour": 81.5376568474138,
     "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": 1.0858605960002023,
-      "bytes": 2526198,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.2877745800014964,
       "bytes": 5791906,
@@ -562,6 +556,12 @@
       "bytes": 6267595,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.9623302439977124,
+      "bytes": 6111937,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -172,3 +172,4 @@
 {"step": 457400, "loss": 0.6224752068519592, "lr": 0.0003, "grad_norm": 0.1910308007215812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66949595786, "tokens_per_second": 152821.2799298036, "data_wait_s": 1.3413999113254249e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 457500, "loss": 0.5915499925613403, "lr": 0.0003, "grad_norm": 0.17881144806741683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66964305240, "tokens_per_second": 152904.36875887544, "data_wait_s": 9.677998605184257e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 457600, "loss": 0.5711995363235474, "lr": 0.0003, "grad_norm": 0.13963287596159765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66978781488, "tokens_per_second": 150979.49881917232, "data_wait_s": 1.1741998605430126e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 457700, "loss": 0.45534321665763855, "lr": 0.0003, "grad_norm": 0.22608440001065322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66993292703, "tokens_per_second": 151032.16086570555, "data_wait_s": 1.5909001376712695e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cb917665f88a8ca9b933db45e6e151185196e983da221c7bd31f2ff34311a6fd
-size 371046
+oid sha256:57a7b088379d2122fb939e38cad88b700328d2a0649beaaf639d039c9173d24c
+size 373164