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": 77175.19553171426,
-      "required_pages_per_second_upper_estimate": 0.0053391590980088124
+      "required_token_positions_per_second": 77093.90372748765,
+      "required_pages_per_second_upper_estimate": 0.005333535142369442
     },
     {
       "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": 9646.899441464282,
-      "required_pages_per_second_upper_estimate": 0.004038955345973095
+      "required_token_positions_per_second": 9636.737965935956,
+      "required_pages_per_second_upper_estimate": 0.004034700948365123
     },
     {
       "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": 19293.798882928564,
-      "required_pages_per_second_upper_estimate": 0.004018389495621373
+      "required_token_positions_per_second": 19273.47593187191,
+      "required_pages_per_second_upper_estimate": 0.004014156760868533
     },
     {
       "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": 28940.698324392848,
-      "required_pages_per_second_upper_estimate": 0.0017349506746596874
+      "required_token_positions_per_second": 28910.21389780787,
+      "required_pages_per_second_upper_estimate": 0.0017331231798329422
     },
     {
       "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": 28940.698324392848,
-      "required_pages_per_second_upper_estimate": 0.0022500119980427427
+      "required_token_positions_per_second": 28910.21389780787,
+      "required_pages_per_second_upper_estimate": 0.0022476419679625824
     },
     {
       "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": 28940.698324392848,
-      "required_pages_per_second_upper_estimate": 0.37122020400447464
+      "required_token_positions_per_second": 28910.21389780787,
+      "required_pages_per_second_upper_estimate": 0.37082918251186964
     }
   ],
   "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": 0.23301981753130008,
-    "producer_p95_seconds": 0.2256024759990396,
-    "gpu_update_seconds": 0.955332856522564,
-    "producer_batches_per_second": 4.291480486914708,
-    "gpu_batches_per_second": 1.0467555817561072,
-    "producer_headroom": 4.0997923122750715,
+    "producer_batch_seconds": 0.08371173818746058,
+    "producer_p95_seconds": 0.2150271819991758,
+    "gpu_update_seconds": 0.9563402089562687,
+    "producer_batches_per_second": 11.945756015251309,
+    "gpu_batches_per_second": 1.0456529910954813,
+    "producer_headroom": 11.42420680376604,
     "sustainable": true,
-    "buffer_seconds": 60.185969960921526,
+    "buffer_seconds": 60.24943316424493,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 294466.42272832594,
-    "upload_bytes_per_second": 5099275.012007909,
+    "generation_bytes_per_second": 324208.36801466544,
+    "upload_bytes_per_second": 5112223.346438631,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.185639490006764,
+    "estimated_full_shards_per_hour": 69.5675685914037,
     "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": 1.1752439570000206,
-      "bytes": 5996670,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1166780060002566,
       "bytes": 4727222,
@@ -562,6 +556,12 @@
       "bytes": 11620692,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.9693775400010054,
+      "bytes": 6039517,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -112,3 +112,4 @@
 {"step": 408300, "loss": 0.5720790028572083, "lr": 0.0003, "grad_norm": 0.15951854378604843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59764349486, "tokens_per_second": 151496.766827079, "data_wait_s": 8.944000001065433e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 408400, "loss": 0.5570839643478394, "lr": 0.0003, "grad_norm": 0.16223706531996163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59779045275, "tokens_per_second": 152563.07967184033, "data_wait_s": 1.2815999070880935e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 408500, "loss": 0.6325985193252563, "lr": 0.0003, "grad_norm": 0.15045436630302347, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59793643857, "tokens_per_second": 151678.50316357208, "data_wait_s": 1.1547001122380607e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 408600, "loss": 0.5965887904167175, "lr": 0.0003, "grad_norm": 0.16157135267235148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59808262489, "tokens_per_second": 152413.77060938033, "data_wait_s": 1.0130999726243317e-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 sha256:16dc3ee34b63be45373684f5e4190cfd1716ea8b6cb3b80a254292ad684a2176
-size 243183
+oid sha256:401abb5be4961c6033c94ba61a39ecead2de8efdef839b49bfc84465b282a559
+size 245301