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": 76119.77437204753,
-      "required_pages_per_second_upper_estimate": 0.006597869118764729
+      "required_token_positions_per_second": 75285.37422551974,
+      "required_pages_per_second_upper_estimate": 0.006525545428831538
     },
     {
       "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": 9514.97179650594,
-      "required_pages_per_second_upper_estimate": 0.003795664511132097
+      "required_token_positions_per_second": 9410.671778189968,
+      "required_pages_per_second_upper_estimate": 0.0037540576744016147
     },
     {
       "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": 19029.94359301188,
-      "required_pages_per_second_upper_estimate": 0.0044022425346748485
+      "required_token_positions_per_second": 18821.343556379936,
+      "required_pages_per_second_upper_estimate": 0.004353986587435304
     },
     {
       "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": 28544.915389517824,
-      "required_pages_per_second_upper_estimate": 0.0017382355836249273
+      "required_token_positions_per_second": 28232.015334569904,
+      "required_pages_per_second_upper_estimate": 0.0017191816119383131
     },
     {
       "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": 28544.915389517824,
-      "required_pages_per_second_upper_estimate": 0.0021974666040682177
+      "required_token_positions_per_second": 28232.015334569904,
+      "required_pages_per_second_upper_estimate": 0.002173378691675538
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28544.915389517824,
-      "required_pages_per_second_upper_estimate": 0.35443231545150455
+      "required_token_positions_per_second": 28232.015334569904,
+      "required_pages_per_second_upper_estimate": 0.35054714397817033
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,27 +158,34 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.05155565389844341,
-    "producer_p95_seconds": 0.1312729589999435,
-    "gpu_update_seconds": 0.9685788037106186,
-    "producer_batches_per_second": 19.396514725035665,
-    "gpu_batches_per_second": 1.032440516113926,
-    "producer_headroom": 18.78705302853044,
+    "producer_batch_seconds": 0.08095469775000197,
+    "producer_p95_seconds": 0.14915560899999036,
+    "gpu_update_seconds": 0.9793137214028507,
+    "producer_batches_per_second": 12.352587654494402,
+    "gpu_batches_per_second": 1.0211232398209602,
+    "producer_headroom": 12.097058584877823,
     "sustainable": true,
-    "buffer_seconds": 60.05188583005835,
+    "buffer_seconds": 60.71745072697674,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "generation_bytes_per_second": 349620.01143116987,
+    "upload_bytes_per_second": 58975.920694439636,
+    "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 75.0203157158024,
     "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": [],
+  "upload_measurements": [
+    {
+      "seconds": 0.8837165979998645,
+      "bytes": 52118,
+      "pages": 1,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -24,3 +24,4 @@
 {"step": 357400, "loss": 0.6018977463245392, "lr": 0.0003, "grad_norm": 0.1774761756098266, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52313238998, "tokens_per_second": 150203.83542135483, "data_wait_s": 1.2816999969800236e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 357500, "loss": 0.6066079773008823, "lr": 0.0003, "grad_norm": 0.1540605306034135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52327985290, "tokens_per_second": 151886.11674555767, "data_wait_s": 1.3465999927575467e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 357600, "loss": 0.5449966806918383, "lr": 0.0003, "grad_norm": 0.16280325971500753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52342744877, "tokens_per_second": 150667.2497874788, "data_wait_s": 1.5768000139360083e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 357700, "loss": 0.6157439351081848, "lr": 0.0003, "grad_norm": 0.1585160787711768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52357514119, "tokens_per_second": 151091.1419066148, "data_wait_s": 1.1921999885089463e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:43a3cd471ec875a0fcdbcaa43bb5a887ed8e25bd958533fffa21e37fd835e1d2
-size 55233
+oid sha256:8a73e162c09fc15e41f881704bc88bd12f498c11c307ef6e406c52badb836d93
+size 57351