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": 75508.17847187209,
-      "required_pages_per_second_upper_estimate": 0.006544857536213159
+      "required_token_positions_per_second": 75645.33855517661,
+      "required_pages_per_second_upper_estimate": 0.006556746224604943
     },
     {
       "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": 9438.522308984011,
-      "required_pages_per_second_upper_estimate": 0.0037651676675379015
+      "required_token_positions_per_second": 9455.667319397076,
+      "required_pages_per_second_upper_estimate": 0.0037720070685324223
     },
     {
       "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": 18877.044617968022,
-      "required_pages_per_second_upper_estimate": 0.004366872047728506
+      "required_token_positions_per_second": 18911.334638794153,
+      "required_pages_per_second_upper_estimate": 0.0043748044405628765
     },
     {
       "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": 28315.566926952033,
-      "required_pages_per_second_upper_estimate": 0.001724269465553058
+      "required_token_positions_per_second": 28367.00195819123,
+      "required_pages_per_second_upper_estimate": 0.001727401589096772
     },
     {
       "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": 28315.566926952033,
-      "required_pages_per_second_upper_estimate": 0.0021798107245427248
+      "required_token_positions_per_second": 28367.00195819123,
+      "required_pages_per_second_upper_estimate": 0.0021837703356287976
     },
     {
       "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": 28315.566926952033,
-      "required_pages_per_second_upper_estimate": 0.3515845751263647
+      "required_token_positions_per_second": 28367.00195819123,
+      "required_pages_per_second_upper_estimate": 0.35222322607237955
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.050682581375003366,
-    "producer_p95_seconds": 0.13704574799999136,
-    "gpu_update_seconds": 0.9764240310400915,
-    "producer_batches_per_second": 19.730644589724857,
-    "gpu_batches_per_second": 1.0241452158185775,
-    "producer_headroom": 19.265475525318518,
+    "producer_batch_seconds": 0.08588451295312183,
+    "producer_p95_seconds": 0.132505854999863,
+    "gpu_update_seconds": 0.9746535795622345,
+    "producer_batches_per_second": 11.64354277174312,
+    "gpu_batches_per_second": 1.0260055685109675,
+    "producer_headroom": 11.348420641265413,
     "sustainable": true,
-    "buffer_seconds": 60.53828992448567,
+    "buffer_seconds": 60.428521932858544,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 143,
+    "upload_pages_per_commit": 142,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 756727.9281404478,
-    "upload_bytes_per_second": 4873718.004538263,
+    "generation_bytes_per_second": 768810.9648677881,
+    "upload_bytes_per_second": 4708605.836080916,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.18809882717169,
+    "estimated_full_shards_per_hour": 82.4844680286657,
     "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": [
@@ -244,6 +244,12 @@
       "bytes": 11637618,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 0.9964154829999643,
+      "bytes": 2622281,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -7,3 +7,4 @@
 {"step": 700, "loss": 1.4169969446957111, "lr": 0.00020999999999999998, "grad_norm": 1.1299403160160901, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 103081101, "tokens_per_second": 149879.6582243547, "data_wait_s": 1.00040001598245e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 800, "loss": 1.3345604501664639, "lr": 0.00023999999999999998, "grad_norm": 0.8847908940671446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 117723562, "tokens_per_second": 148830.22167264167, "data_wait_s": 1.11660001493874e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 900, "loss": 1.2485260888934135, "lr": 0.00026999999999999995, "grad_norm": 0.5126180461192602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 132386398, "tokens_per_second": 148764.82655347895, "data_wait_s": 2.0052000081705046e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 1000, "loss": 1.1477682627737522, "lr": 0.0003, "grad_norm": 0.40807168767980584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 146947989, "tokens_per_second": 149593.43493141377, "data_wait_s": 1.4937000059944694e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a5c3bb1cd0e2b3732a29675af263bcb112e02bb38019c04641ef3124749f57ce
-size 18873
+oid sha256:91aab8257a64f0356dcef4bb0ce36d08765a7829399bd6a3317fbf7d902a8be1
+size 20956