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": 76528.87727013993,
-      "required_pages_per_second_upper_estimate": 0.0052944453010081195
+      "required_token_positions_per_second": 76800.23240009906,
+      "required_pages_per_second_upper_estimate": 0.005313218278529341
     },
     {
       "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": 9566.109658767491,
-      "required_pages_per_second_upper_estimate": 0.004005130351040455
+      "required_token_positions_per_second": 9600.029050012383,
+      "required_pages_per_second_upper_estimate": 0.004019331691837257
     },
     {
       "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": 19132.219317534982,
-      "required_pages_per_second_upper_estimate": 0.003984736733289578
+      "required_token_positions_per_second": 19200.058100024766,
+      "required_pages_per_second_upper_estimate": 0.003998865762653199
     },
     {
       "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": 28698.328976302473,
-      "required_pages_per_second_upper_estimate": 0.0017204210023182359
+      "required_token_positions_per_second": 28800.087150037147,
+      "required_pages_per_second_upper_estimate": 0.0017265212494578948
     },
     {
       "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": 28698.328976302473,
-      "required_pages_per_second_upper_estimate": 0.0022311688472988127
+      "required_token_positions_per_second": 28800.087150037147,
+      "required_pages_per_second_upper_estimate": 0.0022390800977197794
     },
     {
       "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": 28698.328976302473,
-      "required_pages_per_second_upper_estimate": 0.368111350243102
+      "required_token_positions_per_second": 28800.087150037147,
+      "required_pages_per_second_upper_estimate": 0.36941659483635597
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.22266033519537132,
-    "producer_p95_seconds": 0.22427242399862735,
-    "gpu_update_seconds": 0.9634010406261013,
-    "producer_batches_per_second": 4.491145668682116,
-    "gpu_batches_per_second": 1.0379893292933475,
-    "producer_headroom": 4.326774410811758,
+    "producer_batch_seconds": 0.48750585096097154,
+    "producer_p95_seconds": 0.2334124780009006,
+    "gpu_update_seconds": 0.9599970950075524,
+    "producer_batches_per_second": 2.051257432149378,
+    "gpu_batches_per_second": 1.0416698187947464,
+    "producer_headroom": 1.9692011759760546,
     "sustainable": true,
-    "buffer_seconds": 60.69426555944438,
+    "buffer_seconds": 60.4798169854758,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 351504.039535887,
-    "upload_bytes_per_second": 3951323.5074623087,
+    "generation_bytes_per_second": 307152.66115649795,
+    "upload_bytes_per_second": 3891138.9409143873,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.42458428914507,
+    "estimated_full_shards_per_hour": 65.90781093617633,
     "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": [
@@ -538,6 +538,12 @@
       "bytes": 13232606,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.720672621000631,
+      "bytes": 611156,
+      "pages": 21,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -113,3 +113,4 @@
 {"step": 811300, "loss": 0.5379911065101624, "lr": 3e-05, "grad_norm": 0.26333461520991847, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118751270340, "tokens_per_second": 152233.77071425127, "data_wait_s": 8.107001121970825e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 811400, "loss": 0.5185272693634033, "lr": 3e-05, "grad_norm": 0.26407299973044907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118766056365, "tokens_per_second": 152727.44152252012, "data_wait_s": 2.1675999960280024e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 811500, "loss": 0.5479552745819092, "lr": 3e-05, "grad_norm": 0.27058179124603576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118780566330, "tokens_per_second": 138780.58192060035, "data_wait_s": 2.0338000467745587e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 811600, "loss": 0.521442174911499, "lr": 3e-05, "grad_norm": 0.27802499530849445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118795037952, "tokens_per_second": 149513.28307379616, "data_wait_s": 1.6496000171173364e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5f9abf42bde8e72c965a51ca69c2a0693ee29f9eb03610d90d36270fdf5920e0
-size 245301
+oid sha256:4c258f0c7ffdd5e33065a880b3a1b759098708bef67935685584af10380d281a
+size 247419