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": 76697.96010481361,
-      "required_pages_per_second_upper_estimate": 0.005306142843837074
+      "required_token_positions_per_second": 76639.76533724267,
+      "required_pages_per_second_upper_estimate": 0.005302116794786052
     },
     {
       "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": 9587.245013101701,
-      "required_pages_per_second_upper_estimate": 0.004013979282543803
+      "required_token_positions_per_second": 9579.970667155334,
+      "required_pages_per_second_upper_estimate": 0.004010933665801676
     },
     {
       "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": 19174.490026203403,
-      "required_pages_per_second_upper_estimate": 0.003993540607151659
+      "required_token_positions_per_second": 19159.941334310668,
+      "required_pages_per_second_upper_estimate": 0.003990510498305208
     },
     {
       "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": 28761.735039305106,
-      "required_pages_per_second_upper_estimate": 0.0017242220989797912
+      "required_token_positions_per_second": 28739.912001466,
+      "required_pages_per_second_upper_estimate": 0.0017229138411831862
     },
     {
       "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": 28761.735039305106,
-      "required_pages_per_second_upper_estimate": 0.0022360983898034705
+      "required_token_positions_per_second": 28739.912001466,
+      "required_pages_per_second_upper_estimate": 0.002234401744600879
     },
     {
       "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": 28761.735039305106,
-      "required_pages_per_second_upper_estimate": 0.3689246551391735
+      "required_token_positions_per_second": 28739.912001466,
+      "required_pages_per_second_upper_estimate": 0.3686447326415259
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08992381072675926,
-    "producer_p95_seconds": 0.3307782799965935,
-    "gpu_update_seconds": 0.9612771956287373,
-    "producer_batches_per_second": 11.120525163669727,
-    "gpu_batches_per_second": 1.0402826620119034,
-    "producer_headroom": 10.68990724325124,
+    "producer_batch_seconds": 0.2142487486950131,
+    "producer_p95_seconds": 0.3454566420041374,
+    "gpu_update_seconds": 0.9620071209191489,
+    "producer_batches_per_second": 4.667471833982647,
+    "gpu_batches_per_second": 1.039493344960431,
+    "producer_headroom": 4.490141140980866,
     "sustainable": true,
-    "buffer_seconds": 60.560463324610446,
+    "buffer_seconds": 60.60644861790638,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 397111.7455124524,
-    "upload_bytes_per_second": 5416356.790234443,
+    "generation_bytes_per_second": 404315.30007759656,
+    "upload_bytes_per_second": 5530254.389316219,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.21093629865817,
+    "estimated_full_shards_per_hour": 86.7566514181702,
     "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": 1.4434572989994194,
-      "bytes": 5751342,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.5010855370019272,
       "bytes": 6022741,
@@ -562,6 +556,12 @@
       "bytes": 10401411,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.1967187990012462,
+      "bytes": 13416312,
+      "pages": 42,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -341,3 +341,4 @@
 {"step": 679100, "loss": 0.5356208086013794, "lr": 0.0001, "grad_norm": 0.20511929759650416, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99402851670, "tokens_per_second": 151488.33874070636, "data_wait_s": 1.654700463404879e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 679200, "loss": 0.6018109321594238, "lr": 0.0001, "grad_norm": 0.21745714831770144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99417364559, "tokens_per_second": 150898.36260275112, "data_wait_s": 1.4440003724303097e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 679300, "loss": 0.5146587491035461, "lr": 0.0001, "grad_norm": 0.1973563229490574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99431945721, "tokens_per_second": 151347.31785791268, "data_wait_s": 1.2218995834700763e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 679400, "loss": 0.6518879532814026, "lr": 0.0001, "grad_norm": 0.2733330479010058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99446621460, "tokens_per_second": 153069.31936941762, "data_wait_s": 1.0310999641660601e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1f4ab8262dd154986a4630de4d1376ad1aa016b897e615bca56e9a843993b7c9
-size 731337
+oid sha256:d84d9f1aa4e9bff91bca2275da42c8fb4f850e29bd963f254464815dd449349b
+size 733455