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": 77377.6895665996,
-      "required_pages_per_second_upper_estimate": 0.005353168105193074
+      "required_token_positions_per_second": 76760.99976366253,
+      "required_pages_per_second_upper_estimate": 0.005310504073708403
     },
     {
       "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": 9672.21119582495,
-      "required_pages_per_second_upper_estimate": 0.00404955284895437
+      "required_token_positions_per_second": 9595.124970457817,
+      "required_pages_per_second_upper_estimate": 0.004017278456136587
     },
     {
       "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": 19344.4223916499,
-      "required_pages_per_second_upper_estimate": 0.004028933037456325
+      "required_token_positions_per_second": 19190.249940915633,
+      "required_pages_per_second_upper_estimate": 0.003996822981769317
     },
     {
       "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": 29016.63358747485,
-      "required_pages_per_second_upper_estimate": 0.0017395028777350203
+      "required_token_positions_per_second": 28785.37491137345,
+      "required_pages_per_second_upper_estimate": 0.0017256392732142938
     },
     {
       "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": 29016.63358747485,
-      "required_pages_per_second_upper_estimate": 0.0022559156307434447
+      "required_token_positions_per_second": 28785.37491137345,
+      "required_pages_per_second_upper_estimate": 0.002237936285875976
     },
     {
       "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": 29016.63358747485,
-      "required_pages_per_second_upper_estimate": 0.37219422002635744
+      "required_token_positions_per_second": 28785.37491137345,
+      "required_pages_per_second_upper_estimate": 0.3692278820355492
     }
   ],
   "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": 0.1566039216170907,
-    "producer_p95_seconds": 0.3141943459995673,
-    "gpu_update_seconds": 0.9528327921518219,
-    "producer_batches_per_second": 6.385536132646035,
-    "gpu_batches_per_second": 1.0495020828803114,
-    "producer_headroom": 6.084348222655468,
+    "producer_batch_seconds": 0.3489696182192574,
+    "producer_p95_seconds": 0.32621850600116886,
+    "gpu_update_seconds": 0.960487750641592,
+    "producer_batches_per_second": 2.865578972470035,
+    "gpu_batches_per_second": 1.041137692106968,
+    "producer_headroom": 2.7523535015535883,
     "sustainable": true,
-    "buffer_seconds": 60.02846590556478,
+    "buffer_seconds": 60.510728290420296,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348274.0647871874,
-    "upload_bytes_per_second": 5038105.548694799,
+    "generation_bytes_per_second": 343479.7487578489,
+    "upload_bytes_per_second": 5116333.810621351,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.73150689803806,
+    "estimated_full_shards_per_hour": 73.70275828410722,
     "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": 1.4285129730014887,
-      "bytes": 6270013,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.485799860998668,
       "bytes": 7350955,
@@ -562,6 +556,12 @@
       "bytes": 7474922,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.0820681480036,
+      "bytes": 10987528,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -354,3 +354,4 @@
 {"step": 475600, "loss": 0.6041150689125061, "lr": 0.0003, "grad_norm": 0.15503440650720524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69612531171, "tokens_per_second": 152444.62641148645, "data_wait_s": 1.1541997082531452e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 475700, "loss": 0.5949307680130005, "lr": 0.0003, "grad_norm": 0.14134309732396333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69627255935, "tokens_per_second": 153737.3246875719, "data_wait_s": 2.1879000996705145e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 475800, "loss": 0.6057299375534058, "lr": 0.0003, "grad_norm": 0.1476927660982085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69641974503, "tokens_per_second": 153646.5589215426, "data_wait_s": 1.682799484115094e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 475900, "loss": 0.6367251873016357, "lr": 0.0003, "grad_norm": 0.1925758467650204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69656760288, "tokens_per_second": 153248.00235855463, "data_wait_s": 1.4907003787811846e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:feef9b59798fa55f889ddaa4ba16f206bcc21d063c0dac4d4ed3b43f830e60ec
-size 759654
+oid sha256:3d9345b41d86464973cab46044aa0fdcc3909c0320c7bcbac20419cf5db14cf4
+size 761772