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": 76135.57025623454,
-      "required_pages_per_second_upper_estimate": 0.005267235409188141
+      "required_token_positions_per_second": 76852.5117669682,
+      "required_pages_per_second_upper_estimate": 0.00531683508643419
     },
     {
       "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": 9516.946282029317,
-      "required_pages_per_second_upper_estimate": 0.003984546671848233
+      "required_token_positions_per_second": 9606.563970871024,
+      "required_pages_per_second_upper_estimate": 0.004022067726736105
     },
     {
       "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": 19033.892564058635,
-      "required_pages_per_second_upper_estimate": 0.00396425786359132
+      "required_token_positions_per_second": 19213.12794174205,
+      "required_pages_per_second_upper_estimate": 0.004001587866008004
     },
     {
       "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": 28550.83884608795,
-      "required_pages_per_second_upper_estimate": 0.0017115791942162643
+      "required_token_positions_per_second": 28819.69191261307,
+      "required_pages_per_second_upper_estimate": 0.0017276965250395823
     },
     {
       "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": 28550.83884608795,
-      "required_pages_per_second_upper_estimate": 0.0022197021384151553
+      "required_token_positions_per_second": 28819.69191261307,
+      "required_pages_per_second_upper_estimate": 0.0022406042817778205
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28550.83884608795,
-      "required_pages_per_second_upper_estimate": 0.3662195052152737
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28819.69191261307,
+      "required_pages_per_second_upper_estimate": 0.36966806368072586
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.7077038561405686,
-    "producer_p95_seconds": 0.2685609510008362,
-    "gpu_update_seconds": 0.9683778521900887,
-    "producer_batches_per_second": 1.4130204199443752,
-    "gpu_batches_per_second": 1.0326547615049173,
-    "producer_headroom": 1.3683376793664712,
+    "producer_batch_seconds": 0.10130550310145736,
+    "producer_p95_seconds": 0.23146640300183208,
+    "gpu_update_seconds": 0.9593440514157517,
+    "producer_batches_per_second": 9.871132064745792,
+    "gpu_batches_per_second": 1.0423789030893038,
+    "producer_headroom": 9.469811827053162,
     "sustainable": true,
-    "buffer_seconds": 60.0394268357855,
+    "buffer_seconds": 60.43867523919236,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 312571.48582373554,
-    "upload_bytes_per_second": 3850433.1810842073,
+    "generation_bytes_per_second": 312619.9926494732,
+    "upload_bytes_per_second": 3852250.6962522296,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.07056456598329,
+    "estimated_full_shards_per_hour": 67.08097300160549,
     "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": 1.3820675099996151,
-      "bytes": 1056686,
-      "pages": 32,
-      "shards": 1
-    },
     {
       "seconds": 1.7487962389986933,
       "bytes": 7557748,
@@ -562,6 +556,12 @@
       "bytes": 16842562,
       "pages": 44,
       "shards": 5
+    },
+    {
+      "seconds": 1.6551350050031033,
+      "bytes": 2311821,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -173,3 +173,4 @@
 {"step": 817300, "loss": 0.48066920042037964, "lr": 3e-05, "grad_norm": 0.2902877071253834, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119629303320, "tokens_per_second": 150999.28054310384, "data_wait_s": 1.003400029730983e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 817400, "loss": 0.5275933146476746, "lr": 3e-05, "grad_norm": 0.28932289551674817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119644009460, "tokens_per_second": 150913.19473574357, "data_wait_s": 1.6582998796366155e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 817500, "loss": 0.5398319363594055, "lr": 3e-05, "grad_norm": 0.2947766471505412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119658677538, "tokens_per_second": 151747.2772484756, "data_wait_s": 1.1174000974278897e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 817600, "loss": 0.514951765537262, "lr": 3e-05, "grad_norm": 0.2722887812144312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119673341917, "tokens_per_second": 151635.2995165697, "data_wait_s": 8.19400156615302e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:27d216897f1d2da5e43de4b760243ac572e324fb4d912194ac7bcefdd3ebf74f
-size 373164
+oid sha256:e35e00637dd5802e888fd41a5e139327ddd0b8844b216d9b4f199d27ead3ec17
+size 375282