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": 77562.48843468537,
-      "required_pages_per_second_upper_estimate": 0.005365952919679697
+      "required_token_positions_per_second": 77556.015116188,
+      "required_pages_per_second_upper_estimate": 0.005365505080485881
     },
     {
       "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": 9695.31105433567,
-      "required_pages_per_second_upper_estimate": 0.004059224277332909
+      "required_token_positions_per_second": 9694.5018895235,
+      "required_pages_per_second_upper_estimate": 0.004058885496923337
     },
     {
       "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": 19390.62210867134,
-      "required_pages_per_second_upper_estimate": 0.004038555220143422
+      "required_token_positions_per_second": 19389.003779047,
+      "required_pages_per_second_upper_estimate": 0.004038218164760894
     },
     {
       "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": 29085.933163007012,
-      "required_pages_per_second_upper_estimate": 0.0017436572814738484
+      "required_token_positions_per_second": 29083.505668570502,
+      "required_pages_per_second_upper_estimate": 0.001743511756888949
     },
     {
       "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": 29085.933163007012,
-      "required_pages_per_second_upper_estimate": 0.0022613033679244514
+      "required_token_positions_per_second": 29083.505668570502,
+      "required_pages_per_second_upper_estimate": 0.0022611146409094326
     },
     {
       "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": 29085.933163007012,
-      "required_pages_per_second_upper_estimate": 0.373083120573197
+      "required_token_positions_per_second": 29083.505668570502,
+      "required_pages_per_second_upper_estimate": 0.3730519832810059
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,27 +158,34 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.04664811298437144,
-    "producer_p95_seconds": 0.12120914899969648,
-    "gpu_update_seconds": 0.9505625913754128,
-    "producer_batches_per_second": 21.43709436510392,
-    "gpu_batches_per_second": 1.0520085779444086,
-    "producer_headroom": 20.37729997125244,
+    "producer_batch_seconds": 0.04598613185938305,
+    "producer_p95_seconds": 0.1276059299998451,
+    "gpu_update_seconds": 0.9506419313775574,
+    "producer_batches_per_second": 21.74568635295989,
+    "gpu_batches_per_second": 1.0519207779430881,
+    "producer_headroom": 20.672361273708383,
     "sustainable": true,
-    "buffer_seconds": 60.83600584802642,
+    "buffer_seconds": 60.84108360816367,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 71,
+    "upload_pages_per_commit": 70,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "generation_bytes_per_second": 286201.09277528943,
+    "upload_bytes_per_second": 23618.971052202472,
+    "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 61.41209208911908,
     "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": 1.0629590909998115,
+      "bytes": 25106,
+      "pages": 1,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -15,3 +15,4 @@
 {"step": 526500, "loss": 0.5814909338951111, "lr": 0.0003, "grad_norm": 0.17019092939085942, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77062021391, "tokens_per_second": 153103.6932522806, "data_wait_s": 1.4146000012260629e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 526600, "loss": 0.5899880528450012, "lr": 0.0003, "grad_norm": 0.15871447608584385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77076707540, "tokens_per_second": 154316.9364779082, "data_wait_s": 1.3916999705543276e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 526700, "loss": 0.60279381275177, "lr": 0.0003, "grad_norm": 0.201481649817476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77091369057, "tokens_per_second": 153328.03270316366, "data_wait_s": 1.0875000043597538e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 526800, "loss": 0.6574923992156982, "lr": 0.0003, "grad_norm": 0.17117518403959395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77106153896, "tokens_per_second": 153519.46429840563, "data_wait_s": 1.0234000001219101e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:28c79dcf60e270146587f9e5948a356092f2ffd3c28549e45713872440ffbcc0
-size 36171
+oid sha256:728433d06fd832a19151f641cb9d98a919887f50f5240cdbb197b3456edef77f
+size 38289