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": 76850.46589855145,
-      "required_pages_per_second_upper_estimate": 0.0053166935485100516
+      "required_token_positions_per_second": 76980.94850757017,
+      "required_pages_per_second_upper_estimate": 0.005325720638163332
     },
     {
       "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": 9606.308237318932,
-      "required_pages_per_second_upper_estimate": 0.0040219606564381675
+      "required_token_positions_per_second": 9622.61856344627,
+      "required_pages_per_second_upper_estimate": 0.004028789449389344
     },
     {
       "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": 19212.616474637864,
-      "required_pages_per_second_upper_estimate": 0.004001481340898497
+      "required_token_positions_per_second": 19245.23712689254,
+      "required_pages_per_second_upper_estimate": 0.004008275362498696
     },
     {
       "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": 28818.924711956795,
-      "required_pages_per_second_upper_estimate": 0.0017276505325316861
+      "required_token_positions_per_second": 28867.855690338816,
+      "required_pages_per_second_upper_estimate": 0.0017305838699724102
     },
     {
       "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": 28818.924711956795,
-      "required_pages_per_second_upper_estimate": 0.002240544635301354
+      "required_token_positions_per_second": 28867.855690338816,
+      "required_pages_per_second_upper_estimate": 0.0022443488036459278
     },
     {
       "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": 28818.924711956795,
-      "required_pages_per_second_upper_estimate": 0.3696582228544631
+      "required_token_positions_per_second": 28867.855690338816,
+      "required_pages_per_second_upper_estimate": 0.3702858569071564
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.13888366764047078,
-    "producer_p95_seconds": 0.29647000600016327,
-    "gpu_update_seconds": 0.9593695905152566,
-    "producer_batches_per_second": 7.200270679693654,
-    "gpu_batches_per_second": 1.0423511542229744,
-    "producer_headroom": 6.907720733576709,
+    "producer_batch_seconds": 0.10432171927342893,
+    "producer_p95_seconds": 0.302914124000381,
+    "gpu_update_seconds": 0.957743460289395,
+    "producer_batches_per_second": 9.585731590360236,
+    "gpu_batches_per_second": 1.0441209378739444,
+    "producer_headroom": 9.18067174275698,
     "sustainable": true,
-    "buffer_seconds": 60.440284202461164,
+    "buffer_seconds": 60.33783799823189,
     "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": 394537.6680694243,
-    "upload_bytes_per_second": 5974648.966416384,
+    "generation_bytes_per_second": 418285.9131914789,
+    "upload_bytes_per_second": 5887065.276673325,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.65859920084044,
+    "estimated_full_shards_per_hour": 89.75441977318071,
     "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": 1.1740434719977202,
-      "bytes": 8468925,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 1.0248841799984802,
       "bytes": 5963815,
@@ -562,6 +556,12 @@
       "bytes": 7627867,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.9741381140011072,
+      "bytes": 986302,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -280,3 +280,4 @@
 {"step": 425100, "loss": 0.5539427399635315, "lr": 0.0003, "grad_norm": 0.16174625622232439, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62223279926, "tokens_per_second": 141672.09914388717, "data_wait_s": 1.3191001926315948e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 425200, "loss": 0.602100670337677, "lr": 0.0003, "grad_norm": 0.15281578308950716, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62237926307, "tokens_per_second": 151587.57524976644, "data_wait_s": 9.35800198931247e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 425300, "loss": 0.6088168621063232, "lr": 0.0003, "grad_norm": 0.1415135881564908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62252428853, "tokens_per_second": 150517.27127679912, "data_wait_s": 1.5074998373165727e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 425400, "loss": 0.5968486070632935, "lr": 0.0003, "grad_norm": 0.19339706772052453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62266926467, "tokens_per_second": 149814.906819925, "data_wait_s": 1.0260999260935932e-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:fcfaf2b2990dc4119709244dda554fcc96147fc405b151d2280a474705bad90e
-size 602139
+oid sha256:d4570e6165193598df57306c891a1d23b8dda6dc813028998270fc5bd03c5c89
+size 604257