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": 77280.73672527664,
-      "required_pages_per_second_upper_estimate": 0.005346460682668766
+      "required_token_positions_per_second": 76836.61709219146,
+      "required_pages_per_second_upper_estimate": 0.00531573545595241
     },
     {
       "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": 9660.09209065958,
-      "required_pages_per_second_upper_estimate": 0.004044478832697323
+      "required_token_positions_per_second": 9604.577136523932,
+      "required_pages_per_second_upper_estimate": 0.0040212358806847965
     },
     {
       "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": 19320.18418131916,
-      "required_pages_per_second_upper_estimate": 0.004023884857449023
+      "required_token_positions_per_second": 19209.154273047865,
+      "required_pages_per_second_upper_estimate": 0.004000760255611697
     },
     {
       "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": 28980.27627197874,
-      "required_pages_per_second_upper_estimate": 0.0017373233121854347
+      "required_token_positions_per_second": 28813.731409571796,
+      "required_pages_per_second_upper_estimate": 0.001727339201983419
     },
     {
       "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": 28980.27627197874,
-      "required_pages_per_second_upper_estimate": 0.0022530890093825522
+      "required_token_positions_per_second": 28813.731409571796,
+      "required_pages_per_second_upper_estimate": 0.0022401408788838436
     },
     {
       "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": 28980.27627197874,
-      "required_pages_per_second_upper_estimate": 0.3717278674206172
+      "required_token_positions_per_second": 28813.731409571796,
+      "required_pages_per_second_upper_estimate": 0.36959160874760194
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.07767530326565009,
-    "producer_p95_seconds": 0.16441176300031657,
-    "gpu_update_seconds": 0.9540281721445517,
-    "producer_batches_per_second": 12.874104869343,
-    "gpu_batches_per_second": 1.0481870758094163,
-    "producer_headroom": 12.282258736496575,
+    "producer_batch_seconds": 0.12867776183597357,
+    "producer_p95_seconds": 0.16910575099973357,
+    "gpu_update_seconds": 0.9595425044746358,
+    "producer_batches_per_second": 7.7713505871722175,
+    "gpu_batches_per_second": 1.0421633177651837,
+    "producer_headroom": 7.456941205565662,
     "sustainable": true,
-    "buffer_seconds": 60.103774845106756,
+    "buffer_seconds": 60.45117778190206,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314445.29108858394,
-    "upload_bytes_per_second": 4959956.239218832,
+    "generation_bytes_per_second": 339555.90420853713,
+    "upload_bytes_per_second": 4905481.775932796,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.47263955586565,
+    "estimated_full_shards_per_hour": 72.86079258625112,
     "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": [
@@ -382,6 +382,12 @@
       "bytes": 6664316,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 0.9569276100000934,
+      "bytes": 2410711,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -39,3 +39,4 @@
 {"step": 401000, "loss": 0.5702412128448486, "lr": 0.0003, "grad_norm": 0.1436486818860728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58695149928, "tokens_per_second": 152054.28080523902, "data_wait_s": 9.996000699175056e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 401100, "loss": 0.5706354379653931, "lr": 0.0003, "grad_norm": 0.13893425313649424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58709726828, "tokens_per_second": 151841.03226773316, "data_wait_s": 9.716999556985684e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 401200, "loss": 0.6189337372779846, "lr": 0.0003, "grad_norm": 0.17205480959495217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58724418532, "tokens_per_second": 152321.89047118434, "data_wait_s": 2.0074000531167258e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 401300, "loss": 0.5392239689826965, "lr": 0.0003, "grad_norm": 0.19113700635905972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58739160213, "tokens_per_second": 153528.80219505506, "data_wait_s": 1.018900002236478e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c7950692e320e59b0797463e2aba749100ce10ca8251c6fc5317b945095cbedc
-size 87786
+oid sha256:7cee87ab35cae31a7f53c0de5964bfaeec6d7c64ad7d69f99e8a39f05ac45191
+size 89904