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": 76058.80911622461,
-      "required_pages_per_second_upper_estimate": 0.006592584804905027
+      "required_token_positions_per_second": 75032.59243034208,
+      "required_pages_per_second_upper_estimate": 0.00650363494349514
     },
     {
       "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": 9507.351139528077,
-      "required_pages_per_second_upper_estimate": 0.0037926245171246518
+      "required_token_positions_per_second": 9379.07405379276,
+      "required_pages_per_second_upper_estimate": 0.0037414528697114887
     },
     {
       "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": 19014.702279056153,
-      "required_pages_per_second_upper_estimate": 0.004398716724929155
+      "required_token_positions_per_second": 18758.14810758552,
+      "required_pages_per_second_upper_estimate": 0.004339367432558629
     },
     {
       "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": 28522.05341858423,
-      "required_pages_per_second_upper_estimate": 0.001736843409542563
+      "required_token_positions_per_second": 28137.22216137828,
+      "required_pages_per_second_upper_estimate": 0.0017134092050322862
     },
     {
       "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": 28522.05341858423,
-      "required_pages_per_second_upper_estimate": 0.0021957066262597636
+      "required_token_positions_per_second": 28137.22216137828,
+      "required_pages_per_second_upper_estimate": 0.0021660812507989487
     },
     {
       "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": 28522.05341858423,
-      "required_pages_per_second_upper_estimate": 0.354148446286604
+      "required_token_positions_per_second": 28137.22216137828,
+      "required_pages_per_second_upper_estimate": 0.34937013001947276
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08809222614073065,
-    "producer_p95_seconds": 0.354377724997903,
-    "gpu_update_seconds": 0.969355172092388,
-    "producer_batches_per_second": 11.351739464529622,
-    "gpu_batches_per_second": 1.0316136219105987,
-    "producer_headroom": 11.003867362187064,
+    "producer_batch_seconds": 0.43456916295312453,
+    "producer_p95_seconds": 0.33482125799855567,
+    "gpu_update_seconds": 0.9826129900609096,
+    "producer_batches_per_second": 2.3011296825676206,
+    "gpu_batches_per_second": 1.0176946672952212,
+    "producer_headroom": 2.2611199179056816,
     "sustainable": true,
-    "buffer_seconds": 60.10002066972805,
+    "buffer_seconds": 60.922005383776394,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 345097.6647442144,
-    "upload_bytes_per_second": 4460975.442273614,
+    "generation_bytes_per_second": 333768.8286494337,
+    "upload_bytes_per_second": 4421273.620485497,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.04992539162468,
+    "estimated_full_shards_per_hour": 71.61902088749177,
     "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": 1.560656753998046,
-      "bytes": 11939208,
-      "pages": 42,
-      "shards": 4
-    },
     {
       "seconds": 1.4417221109979437,
       "bytes": 5916590,
@@ -562,6 +556,12 @@
       "bytes": 2398547,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.0771264450013405,
+      "bytes": 6009796,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -277,3 +277,4 @@
 {"step": 182700, "loss": 0.5470570288598537, "lr": 0.0003, "grad_norm": 0.1333711140535544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26739831087, "tokens_per_second": 151449.40135941628, "data_wait_s": 1.5121000615181401e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 182800, "loss": 0.5964217279106379, "lr": 0.0003, "grad_norm": 0.1507659737046911, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26754375959, "tokens_per_second": 148559.30736765746, "data_wait_s": 1.060299837263301e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 182900, "loss": 0.5815033316612244, "lr": 0.0003, "grad_norm": 0.14769942496852517, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26768939256, "tokens_per_second": 132840.5291579347, "data_wait_s": 11.942335987998376, "gpu_peak_gb": 85.46427917480469}
+{"step": 183000, "loss": 0.6215140540152788, "lr": 0.0003, "grad_norm": 0.13817083573035038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26783605310, "tokens_per_second": 148652.63892421804, "data_wait_s": 1.3806999049847946e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:63da51535c83ad8bd1cfacaaa1bcb766359a21ac493fa159d2d69ac40df862c5
-size 595002
+oid sha256:0c84900ea2f3669f90bbd55ef19b8929b7634a33affc2b9cae0fc54f3130a50c
+size 597120