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": 77103.06102905466,
-      "required_pages_per_second_upper_estimate": 0.005334168665739715
+      "required_token_positions_per_second": 76919.94848440614,
+      "required_pages_per_second_upper_estimate": 0.00532150051502129
     },
     {
       "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": 9637.882628631833,
-      "required_pages_per_second_upper_estimate": 0.004035180194732611
+      "required_token_positions_per_second": 9614.993560550767,
+      "required_pages_per_second_upper_estimate": 0.004025597019905164
     },
     {
       "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": 19275.765257263665,
-      "required_pages_per_second_upper_estimate": 0.004014633566974064
+      "required_token_positions_per_second": 19229.987121101534,
+      "required_pages_per_second_upper_estimate": 0.004005099188462127
     },
     {
       "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": 28913.647885895498,
-      "required_pages_per_second_upper_estimate": 0.0017333290421753993
+      "required_token_positions_per_second": 28844.9806816523,
+      "required_pages_per_second_upper_estimate": 0.0017292125481297696
     },
     {
       "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": 28913.647885895498,
-      "required_pages_per_second_upper_estimate": 0.002247908945431878
+      "required_token_positions_per_second": 28844.9806816523,
+      "required_pages_per_second_upper_estimate": 0.002242570372337081
     },
     {
       "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": 28913.647885895498,
-      "required_pages_per_second_upper_estimate": 0.3708732300239286
+      "required_token_positions_per_second": 28844.9806816523,
+      "required_pages_per_second_upper_estimate": 0.36999244085699645
     }
   ],
   "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": 0.11831319244492988,
-    "producer_p95_seconds": 0.3144909490001737,
-    "gpu_update_seconds": 0.9562266272699234,
-    "producer_batches_per_second": 8.452142819706776,
-    "gpu_batches_per_second": 1.0457771949470305,
-    "producer_headroom": 8.08216402169191,
+    "producer_batch_seconds": 0.10863790405483087,
+    "producer_p95_seconds": 0.3040151659952244,
+    "gpu_update_seconds": 0.9585029820313357,
+    "producer_batches_per_second": 9.20489039898347,
+    "gpu_batches_per_second": 1.0432935721083731,
+    "producer_headroom": 8.822914896697267,
     "sustainable": true,
-    "buffer_seconds": 60.242277518005174,
+    "buffer_seconds": 60.38568786797415,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 355518.7531721854,
-    "upload_bytes_per_second": 5233381.266217245,
+    "generation_bytes_per_second": 356152.7446681171,
+    "upload_bytes_per_second": 5028995.154457731,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.28604837774441,
+    "estimated_full_shards_per_hour": 76.42208819420466,
     "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": 0.9963944320006703,
-      "bytes": 6337215,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.0649026120008784,
       "bytes": 9997467,
@@ -562,6 +556,12 @@
       "bytes": 11745696,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 3.1945914269963396,
+      "bytes": 906980,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -351,3 +351,4 @@
 {"step": 475300, "loss": 0.5609617233276367, "lr": 0.0003, "grad_norm": 0.17542387447329194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69568895568, "tokens_per_second": 153043.73075176837, "data_wait_s": 1.012799475574866e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 475400, "loss": 0.5558681488037109, "lr": 0.0003, "grad_norm": 0.1455002630257173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69583454500, "tokens_per_second": 151211.05860750322, "data_wait_s": 1.4534998626913875e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 475500, "loss": 0.6086797714233398, "lr": 0.0003, "grad_norm": 0.16874298172772706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69597926333, "tokens_per_second": 150589.9090848339, "data_wait_s": 1.1281001206953079e-05, "gpu_peak_gb": 89.37899398803711}
+{"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}

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 sha256:49453e7c25000d47e391fb8afe89467cd07681a630ec41e5418106ee31bd151b
-size 753300
+oid sha256:654326cb7a13c4bd6d387dc1953a6cd60f5aad453899c930ee109cdc81a503d1
+size 755418