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": 77526.30520890247,
-      "required_pages_per_second_upper_estimate": 0.005363449680163386
+      "required_token_positions_per_second": 76986.36874606617,
+      "required_pages_per_second_upper_estimate": 0.005326095622839171
     },
     {
       "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": 9690.788151112809,
-      "required_pages_per_second_upper_estimate": 0.004057330632202457
+      "required_token_positions_per_second": 9623.296093258272,
+      "required_pages_per_second_upper_estimate": 0.004029073116973198
     },
     {
       "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": 19381.576302225618,
-      "required_pages_per_second_upper_estimate": 0.004036671217214735
+      "required_token_positions_per_second": 19246.592186516544,
+      "required_pages_per_second_upper_estimate": 0.004008557585683051
     },
     {
       "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": 29072.36445333843,
-      "required_pages_per_second_upper_estimate": 0.0017428438580474367
+      "required_token_positions_per_second": 28869.888279774816,
+      "required_pages_per_second_upper_estimate": 0.001730705720605514
     },
     {
       "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": 29072.36445333843,
-      "required_pages_per_second_upper_estimate": 0.0022602484604301656
+      "required_token_positions_per_second": 28869.888279774816,
+      "required_pages_per_second_upper_estimate": 0.002244506828534161
     },
     {
       "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": 29072.36445333843,
-      "required_pages_per_second_upper_estimate": 0.37290907573451376
+      "required_token_positions_per_second": 28869.888279774816,
+      "required_pages_per_second_upper_estimate": 0.37031192878201685
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08868935826575353,
-    "producer_p95_seconds": 0.3115197459992487,
-    "gpu_update_seconds": 0.9510062397702618,
-    "producer_batches_per_second": 11.275309908135164,
-    "gpu_batches_per_second": 1.051517811535678,
-    "producer_headroom": 10.722890077979997,
+    "producer_batch_seconds": 0.14881810605459123,
+    "producer_p95_seconds": 0.3455620070017176,
+    "gpu_update_seconds": 0.9576760301968046,
+    "producer_batches_per_second": 6.7196124618947115,
+    "gpu_batches_per_second": 1.0441944545636146,
+    "producer_headroom": 6.435211786968304,
     "sustainable": true,
-    "buffer_seconds": 60.86439934529675,
+    "buffer_seconds": 60.333589902398685,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340181.7727223217,
-    "upload_bytes_per_second": 5134501.118338727,
+    "generation_bytes_per_second": 340166.5371424452,
+    "upload_bytes_per_second": 5010787.708689514,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.99508939983595,
+    "estimated_full_shards_per_hour": 72.99182019906061,
     "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": 1.4335154450018308,
-      "bytes": 16606396,
-      "pages": 45,
-      "shards": 4
-    },
     {
       "seconds": 0.9105693140008952,
       "bytes": 1064335,
@@ -562,6 +556,12 @@
       "bytes": 5521564,
       "pages": 41,
       "shards": 2
+    },
+    {
+      "seconds": 1.7268630670005223,
+      "bytes": 7581932,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -371,3 +371,4 @@
 {"step": 922100, "loss": 0.5193803310394287, "lr": 3e-05, "grad_norm": 0.3367932811834201, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134968918631, "tokens_per_second": 153406.63493892195, "data_wait_s": 9.061994205694646e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 922200, "loss": 0.5152973532676697, "lr": 3e-05, "grad_norm": 0.3198962327742503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134983648474, "tokens_per_second": 153442.0880857307, "data_wait_s": 1.638800313230604e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 922300, "loss": 0.49285024404525757, "lr": 3e-05, "grad_norm": 0.33830968570608894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134998494351, "tokens_per_second": 154051.0488153253, "data_wait_s": 1.481799699831754e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 922400, "loss": 0.5174910426139832, "lr": 3e-05, "grad_norm": 0.3362519461787307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135013189439, "tokens_per_second": 153450.3945819103, "data_wait_s": 1.3343000318855047e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c311cd892a511cd509c8e019b59f89d512bcefa194e2e930d4d688434f27a177
-size 795660
+oid sha256:867c70a9e6305c4c175573292f56a9aacac1b11aa45fc865718220b0f2b5d924
+size 797778