Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74346.91513274066,
-      "required_pages_per_second_upper_estimate": 0.006444202173172272
+      "required_token_positions_per_second": 74745.52749795323,
+      "required_pages_per_second_upper_estimate": 0.0064787528827150945
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9293.364391592582,
-      "required_pages_per_second_upper_estimate": 0.0037072620039861905
+      "required_token_positions_per_second": 9343.190937244153,
+      "required_pages_per_second_upper_estimate": 0.0037271385580198474
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18586.728783185165,
-      "required_pages_per_second_upper_estimate": 0.004299712588735677
+      "required_token_positions_per_second": 18686.381874488306,
+      "required_pages_per_second_upper_estimate": 0.004322765577574153
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27880.093174777743,
-      "required_pages_per_second_upper_estimate": 0.0016977513987998462
+      "required_token_positions_per_second": 28029.57281173246,
+      "required_pages_per_second_upper_estimate": 0.0017068539244313441
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27880.093174777743,
-      "required_pages_per_second_upper_estimate": 0.0021462867496318486
+      "required_token_positions_per_second": 28029.57281173246,
+      "required_pages_per_second_upper_estimate": 0.0021577941058707416
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27880.093174777743,
-      "required_pages_per_second_upper_estimate": 0.3461774485612544
+      "required_token_positions_per_second": 28029.57281173246,
+      "required_pages_per_second_upper_estimate": 0.34803348537606893
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1163612994609764,
-    "producer_p95_seconds": 0.3126731939992169,
-    "gpu_update_seconds": 0.9916753084961814,
-    "producer_batches_per_second": 8.593922589661057,
-    "gpu_batches_per_second": 1.0083945737405144,
-    "producer_headroom": 8.52238083529443,
+    "producer_batch_seconds": 0.06864258017967018,
+    "producer_p95_seconds": 0.22206446399650304,
+    "gpu_update_seconds": 0.9863867774833606,
+    "producer_batches_per_second": 14.568216949050077,
+    "gpu_batches_per_second": 1.0138010999613882,
+    "producer_headroom": 14.369896570051981,
     "sustainable": true,
-    "buffer_seconds": 60.49219381826706,
+    "buffer_seconds": 60.169593426484994,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 225041.05046790154,
-    "upload_bytes_per_second": 5858635.243720461,
+    "generation_bytes_per_second": 254869.45210472695,
+    "upload_bytes_per_second": 5971109.374092571,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 48.28857074287209,
+    "estimated_full_shards_per_hour": 54.68905136448247,
     "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": 0.7505128630000399,
-      "bytes": 1172556,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.1141684859994712,
       "bytes": 8597889,
@@ -562,6 +556,12 @@
       "bytes": 761758,
       "pages": 22,
       "shards": 2
+    },
+    {
+      "seconds": 1.1483951949994662,
+      "bytes": 11735955,
+      "pages": 47,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -190,3 +190,4 @@
 {"step": 19000, "loss": 0.6818721396848559, "lr": 0.0003, "grad_norm": 0.1407577781466548, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2780329225, "tokens_per_second": 148163.24897849787, "data_wait_s": 1.359100133413449e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 19100, "loss": 0.6865865923464298, "lr": 0.0003, "grad_norm": 0.14322243062285178, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2794872890, "tokens_per_second": 147410.04677401582, "data_wait_s": 1.1726999218808487e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 19200, "loss": 0.7024876587092876, "lr": 0.0003, "grad_norm": 0.16395474842740984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2809542028, "tokens_per_second": 147541.79972931175, "data_wait_s": 1.269100175704807e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 19300, "loss": 0.7050697933882475, "lr": 0.0003, "grad_norm": 0.1251860720431679, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2824197435, "tokens_per_second": 149184.74466385331, "data_wait_s": 1.223800063598901e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9050ebfec023b2fb1fbea164a20cf57964cfa38aef6134e591d3f138b9023970
-size 403396
+oid sha256:fd578cbb04299dfe666e48e1d252a947bc23f89114c037feba3f0666d67c5246
+size 405514