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": 75318.78442062503,
-      "required_pages_per_second_upper_estimate": 0.006528441339865899
+      "required_token_positions_per_second": 74618.99625927229,
+      "required_pages_per_second_upper_estimate": 0.006467785475636723
     },
     {
       "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": 9414.848052578129,
-      "required_pages_per_second_upper_estimate": 0.0037557236526959185
+      "required_token_positions_per_second": 9327.374532409036,
+      "required_pages_per_second_upper_estimate": 0.0037208291576547935
     },
     {
       "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": 18829.696105156258,
-      "required_pages_per_second_upper_estimate": 0.0043559188025948685
+      "required_token_positions_per_second": 18654.749064818072,
+      "required_pages_per_second_upper_estimate": 0.00431544788377538
     },
     {
       "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": 28244.544157734388,
-      "required_pages_per_second_upper_estimate": 0.0017199445515353753
+      "required_token_positions_per_second": 27982.123597227106,
+      "required_pages_per_second_upper_estimate": 0.001703964516214757
     },
     {
       "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": 28244.544157734388,
-      "required_pages_per_second_upper_estimate": 0.002174343195164742
+      "required_token_positions_per_second": 27982.123597227106,
+      "required_pages_per_second_upper_estimate": 0.002154141333990287
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28244.544157734388,
-      "required_pages_per_second_upper_estimate": 0.35070271003059944
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27982.123597227106,
+      "required_pages_per_second_upper_estimate": 0.3474443249342179
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.07266573018752354,
-    "producer_p95_seconds": 0.1794149010002002,
-    "gpu_update_seconds": 0.9788793136684052,
-    "producer_batches_per_second": 13.761645240739584,
-    "gpu_batches_per_second": 1.021576394593981,
-    "producer_headroom": 13.470989848203239,
+    "producer_batch_seconds": 0.47532198563281725,
+    "producer_p95_seconds": 0.1731516890004059,
+    "gpu_update_seconds": 0.9880593909870294,
+    "producer_batches_per_second": 2.103837041471279,
+    "gpu_batches_per_second": 1.012084910200633,
+    "producer_headroom": 2.078715945932066,
     "sustainable": true,
-    "buffer_seconds": 60.690517447441124,
+    "buffer_seconds": 60.27162285020879,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430026.7910484539,
-    "upload_bytes_per_second": 5311825.153442362,
+    "generation_bytes_per_second": 366206.74584986235,
+    "upload_bytes_per_second": 5245685.810701095,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.27373884763921,
+    "estimated_full_shards_per_hour": 78.57944280263808,
     "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": [
@@ -448,6 +448,12 @@
       "bytes": 5966016,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.7928353870001956,
+      "bytes": 647903,
+      "pages": 20,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -53,3 +53,4 @@
 {"step": 320300, "loss": 0.5900352392345667, "lr": 0.0003, "grad_norm": 0.1685965670996436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46883153945, "tokens_per_second": 149671.4621128526, "data_wait_s": 1.3650000255438499e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 320400, "loss": 0.5965177156031132, "lr": 0.0003, "grad_norm": 0.16110717642181044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46897838392, "tokens_per_second": 148941.93556122648, "data_wait_s": 1.3484000191965606e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 320500, "loss": 0.6146031823009253, "lr": 0.0003, "grad_norm": 0.16981394066016361, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46912323849, "tokens_per_second": 131657.32953579738, "data_wait_s": 1.104199964174768e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 320600, "loss": 0.5646807532757521, "lr": 0.0003, "grad_norm": 0.14598938220835858, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46926873367, "tokens_per_second": 148443.65597854287, "data_wait_s": 1.2337000043771695e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:440746cbbbb5cc928214dd076c9969e228ff0015a834f2d6a790976c1c28ae6d
-size 117438
+oid sha256:4badfa95a5770fab7b553f7c724b6e659ff5a5b38128edc0d62bd31c5bd0ca10
+size 119556