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": 76674.17478151721,
-      "required_pages_per_second_upper_estimate": 0.005304497319981875
+      "required_token_positions_per_second": 76997.41244358459,
+      "required_pages_per_second_upper_estimate": 0.005326859651458404
     },
     {
       "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": 9584.271847689652,
-      "required_pages_per_second_upper_estimate": 0.004012734480272531
+      "required_token_positions_per_second": 9624.676555448073,
+      "required_pages_per_second_upper_estimate": 0.004029651087664739
     },
     {
       "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": 19168.543695379303,
-      "required_pages_per_second_upper_estimate": 0.003992302143256288
+      "required_token_positions_per_second": 19249.353110896147,
+      "required_pages_per_second_upper_estimate": 0.00400913261342085
     },
     {
       "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": 28752.815543068955,
-      "required_pages_per_second_upper_estimate": 0.0017236873887997162
+      "required_token_positions_per_second": 28874.02966634422,
+      "required_pages_per_second_upper_estimate": 0.0017309539904068188
     },
     {
       "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": 28752.815543068955,
-      "required_pages_per_second_upper_estimate": 0.0022354049382038274
+      "required_token_positions_per_second": 28874.02966634422,
+      "required_pages_per_second_upper_estimate": 0.0022448288031238976
     },
     {
       "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": 28752.815543068955,
-      "required_pages_per_second_upper_estimate": 0.3688102454184651
+      "required_token_positions_per_second": 28874.02966634422,
+      "required_pages_per_second_upper_estimate": 0.37036505004225473
     }
   ],
   "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": 0.09150724757810735,
-    "producer_p95_seconds": 0.22858483500021975,
-    "gpu_update_seconds": 0.9615753962802687,
-    "producer_batches_per_second": 10.928096150487264,
-    "gpu_batches_per_second": 1.039960052917714,
-    "producer_headroom": 10.50818838649367,
+    "producer_batch_seconds": 0.34258388731251443,
+    "producer_p95_seconds": 0.23198293300083606,
+    "gpu_update_seconds": 0.9575386712380749,
+    "producer_batches_per_second": 2.91899309055295,
+    "gpu_batches_per_second": 1.044344244297751,
+    "producer_headroom": 2.795048765281193,
     "sustainable": true,
-    "buffer_seconds": 60.57924996565693,
+    "buffer_seconds": 60.32493628799872,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 404962.9921328912,
-    "upload_bytes_per_second": 4933576.94924735,
+    "generation_bytes_per_second": 381057.9463217083,
+    "upload_bytes_per_second": 4709599.254748465,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.89563105573704,
+    "estimated_full_shards_per_hour": 81.76616470564305,
     "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": 1.0707301010006631,
-      "bytes": 10039688,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.0178890370007139,
       "bytes": 11702648,
@@ -562,6 +556,12 @@
       "bytes": 2485686,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 4.417064211000252,
+      "bytes": 6221614,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -170,3 +170,4 @@
 {"step": 457200, "loss": 0.6129456758499146, "lr": 0.0003, "grad_norm": 0.1519546098089188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66920168103, "tokens_per_second": 152319.61133741678, "data_wait_s": 1.44219993671868e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 457300, "loss": 0.5756521224975586, "lr": 0.0003, "grad_norm": 0.17757718417413582, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66934877632, "tokens_per_second": 152934.34616986127, "data_wait_s": 1.2229000276420265e-05, "gpu_peak_gb": 89.37899398803711}
 {"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}

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:bb92c7bc52fd988282ac4046b0cb846003f3268e51b0303e5c8b2acc5b906c56
-size 366810
+oid sha256:16028bbadc7bea836ba179a55464820fd65291d02fdc7cf80b0886e9e1f4f7f8
+size 368928