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": 75576.96775911974,
-      "required_pages_per_second_upper_estimate": 0.006550820017286931
+      "required_token_positions_per_second": 75556.2576872194,
+      "required_pages_per_second_upper_estimate": 0.006549024920743807
     },
     {
       "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": 9447.120969889967,
-      "required_pages_per_second_upper_estimate": 0.0037685978019347246
+      "required_token_positions_per_second": 9444.532210902426,
+      "required_pages_per_second_upper_estimate": 0.003767565107269198
     },
     {
       "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": 18894.241939779935,
-      "required_pages_per_second_upper_estimate": 0.004370850345467165
+      "required_token_positions_per_second": 18889.06442180485,
+      "required_pages_per_second_upper_estimate": 0.004369652617804834
     },
     {
       "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": 28341.362909669904,
-      "required_pages_per_second_upper_estimate": 0.0017258403055595147
+      "required_token_positions_per_second": 28333.59663270728,
+      "required_pages_per_second_upper_estimate": 0.0017253673800389968
     },
     {
       "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": 28341.362909669904,
-      "required_pages_per_second_upper_estimate": 0.002181796570700197
+      "required_token_positions_per_second": 28333.59663270728,
+      "required_pages_per_second_upper_estimate": 0.0021811987012011836
     },
     {
       "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": 28341.362909669904,
-      "required_pages_per_second_upper_estimate": 0.35190487489812017
+      "required_token_positions_per_second": 28333.59663270728,
+      "required_pages_per_second_upper_estimate": 0.35180844373030135
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.07983603222660918,
-    "producer_p95_seconds": 0.27512342699992587,
-    "gpu_update_seconds": 0.975535301111672,
-    "producer_batches_per_second": 12.525672583045806,
-    "gpu_batches_per_second": 1.0250782302397967,
-    "producer_headroom": 12.219235774927805,
+    "producer_batch_seconds": 0.10101006653127342,
+    "producer_p95_seconds": 0.280451587997959,
+    "gpu_update_seconds": 0.975802696650384,
+    "producer_batches_per_second": 9.900003379271045,
+    "gpu_batches_per_second": 1.0247973319121555,
+    "producer_headroom": 9.6604499943406,
     "sustainable": true,
-    "buffer_seconds": 60.48318866892366,
+    "buffer_seconds": 60.49976719232381,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 372898.2791884782,
-    "upload_bytes_per_second": 4146285.5231759693,
+    "generation_bytes_per_second": 424260.16566595936,
+    "upload_bytes_per_second": 4035375.0496331183,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.01529008618125,
+    "estimated_full_shards_per_hour": 91.0363552807244,
     "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": 1.3280163199997332,
-      "bytes": 13612044,
-      "pages": 47,
-      "shards": 4
-    },
     {
       "seconds": 0.827956348999578,
       "bytes": 1165763,
@@ -562,6 +556,12 @@
       "bytes": 11334677,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.0467089929989015,
+      "bytes": 955823,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -267,3 +267,4 @@
 {"step": 341700, "loss": 0.6026564370840788, "lr": 0.0003, "grad_norm": 0.14756879543224583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50015922048, "tokens_per_second": 149162.0375101352, "data_wait_s": 1.888599945232272e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 341800, "loss": 0.5851886924356222, "lr": 0.0003, "grad_norm": 0.1859805109652498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50030654863, "tokens_per_second": 149067.29383108535, "data_wait_s": 1.384899951517582e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 341900, "loss": 0.66478824429214, "lr": 0.0003, "grad_norm": 0.17863731514469988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50045282204, "tokens_per_second": 148654.6301315183, "data_wait_s": 1.461700230720453e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 342000, "loss": 0.5790649708360434, "lr": 0.0003, "grad_norm": 0.1369144824027103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50059709616, "tokens_per_second": 145901.03387296785, "data_wait_s": 1.6900001355679706e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5fdc0b3b84ade33b2fe95b3501d8b2c94c23c2978da5a7f9e631d9ed6eb3e62e
-size 573822
+oid sha256:2cd23eb7ce8ed53ad44d5dc94749bc3b4e992807623348a0e073047e6e2e56ba
+size 575940