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": 75639.41362274846,
-      "required_pages_per_second_upper_estimate": 0.005232910142041574
+      "required_token_positions_per_second": 76071.75843871386,
+      "required_pages_per_second_upper_estimate": 0.005262820759588245
     },
     {
       "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": 9454.926702843557,
-      "required_pages_per_second_upper_estimate": 0.003958580369159241
+      "required_token_positions_per_second": 9508.969804839233,
+      "required_pages_per_second_upper_estimate": 0.003981207087416529
     },
     {
       "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": 18909.853405687114,
-      "required_pages_per_second_upper_estimate": 0.003938423778035112
+      "required_token_positions_per_second": 19017.939609678466,
+      "required_pages_per_second_upper_estimate": 0.003960935283901628
     },
     {
       "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": 28364.78010853067,
-      "required_pages_per_second_upper_estimate": 0.0017004252569949479
+      "required_token_positions_per_second": 28526.9094145177,
+      "required_pages_per_second_upper_estimate": 0.0017101446613317541
     },
     {
       "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": 28364.78010853067,
-      "required_pages_per_second_upper_estimate": 0.0022052368899559737
+      "required_token_positions_per_second": 28526.9094145177,
+      "required_pages_per_second_upper_estimate": 0.002217841730365014
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28364.78010853067,
-      "required_pages_per_second_upper_estimate": 0.3638329435041966
+      "required_token_positions_per_second": 28526.9094145177,
+      "required_pages_per_second_upper_estimate": 0.3659125641605123
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09000239571867041,
-    "producer_p95_seconds": 0.3179241869947873,
-    "gpu_update_seconds": 0.9747299254290676,
-    "producer_batches_per_second": 11.110815351246883,
-    "gpu_batches_per_second": 1.0259252064717401,
-    "producer_headroom": 10.830044218777013,
+    "producer_batch_seconds": 0.13340013085081637,
+    "producer_p95_seconds": 0.3207498769988888,
+    "gpu_update_seconds": 0.969190163513808,
+    "producer_batches_per_second": 7.496244521066602,
+    "gpu_batches_per_second": 1.0317892583375905,
+    "producer_headroom": 7.265286453112028,
     "sustainable": true,
-    "buffer_seconds": 60.43325537660219,
+    "buffer_seconds": 60.0897901378561,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300293.71124581416,
-    "upload_bytes_per_second": 4889169.235051179,
+    "generation_bytes_per_second": 321212.78327165334,
+    "upload_bytes_per_second": 4817801.860846768,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.43603995352572,
+    "estimated_full_shards_per_hour": 68.92478583919716,
     "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": 1.578546805998485,
-      "bytes": 6985484,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.4114699799974915,
       "bytes": 10209990,
@@ -562,6 +556,12 @@
       "bytes": 5996229,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.6789640829956625,
+      "bytes": 1167744,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -376,3 +376,4 @@
 {"step": 757600, "loss": 0.5609956979751587, "lr": 5e-05, "grad_norm": 0.2824734367028012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110891083389, "tokens_per_second": 151220.20169329786, "data_wait_s": 1.5263001841958612e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 757700, "loss": 0.5231772661209106, "lr": 5e-05, "grad_norm": 0.2378513069707813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110905676625, "tokens_per_second": 150813.0811246651, "data_wait_s": 1.3944001693744212e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 757800, "loss": 0.5440947413444519, "lr": 5e-05, "grad_norm": 0.244408394711463, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110920257797, "tokens_per_second": 150849.8226745286, "data_wait_s": 1.3450000551529229e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 757900, "loss": 0.5424357652664185, "lr": 5e-05, "grad_norm": 0.23454591368320088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110934907123, "tokens_per_second": 151033.32091534746, "data_wait_s": 1.4106000890024006e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:330b94d9d8e0cea110a94ae3b7ee8c1b375bae98f93785f9b179fde6be3b93b2
-size 806250
+oid sha256:a8affdf396ebc215963f07c1a71166ee328004b6e10cb1c79576e56137e3e9f8
+size 808368