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": 77468.18380086037,
-      "required_pages_per_second_upper_estimate": 0.005359428706294786
+      "required_token_positions_per_second": 77284.91310204157,
+      "required_pages_per_second_upper_estimate": 0.005346749614103893
     },
     {
       "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": 9683.522975107546,
-      "required_pages_per_second_upper_estimate": 0.00405428885472318
+      "required_token_positions_per_second": 9660.614137755196,
+      "required_pages_per_second_upper_estimate": 0.004044697402914675
     },
     {
       "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": 19367.04595021509,
-      "required_pages_per_second_upper_estimate": 0.004033644928082094
+      "required_token_positions_per_second": 19321.22827551039,
+      "required_pages_per_second_upper_estimate": 0.004024102314734441
     },
     {
       "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": 29050.568925322637,
-      "required_pages_per_second_upper_estimate": 0.0017415372494227345
+      "required_token_positions_per_second": 28981.842413265585,
+      "required_pages_per_second_upper_estimate": 0.0017374171999642218
     },
     {
       "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": 29050.568925322637,
-      "required_pages_per_second_upper_estimate": 0.0022585539539953336
+      "required_token_positions_per_second": 28981.842413265585,
+      "required_pages_per_second_upper_estimate": 0.002253210769978877
     },
     {
       "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": 29050.568925322637,
-      "required_pages_per_second_upper_estimate": 0.3726295061033419
+      "required_token_positions_per_second": 28981.842413265585,
+      "required_pages_per_second_upper_estimate": 0.3717479561994534
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11562517242219883,
-    "producer_p95_seconds": 0.3326396409975132,
-    "gpu_update_seconds": 0.9517197432887432,
-    "producer_batches_per_second": 8.648635751638546,
-    "gpu_batches_per_second": 1.0507294894864958,
-    "producer_headroom": 8.231077397347283,
+    "producer_batch_seconds": 0.1345064610079021,
+    "producer_p95_seconds": 0.33427594100066926,
+    "gpu_update_seconds": 0.9539766176958074,
+    "producer_batches_per_second": 7.434587100921875,
+    "gpu_batches_per_second": 1.0482437215446174,
+    "producer_headroom": 7.092422256502329,
     "sustainable": true,
-    "buffer_seconds": 60.910063570479565,
+    "buffer_seconds": 60.10052691483587,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 284129.44400817354,
-    "upload_bytes_per_second": 5028264.943100925,
+    "generation_bytes_per_second": 283254.0497448238,
+    "upload_bytes_per_second": 4967050.114951292,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.96756448921113,
+    "estimated_full_shards_per_hour": 60.77972525843177,
     "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": 1.5793520259976503,
-      "bytes": 13424823,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 0.9566230939963134,
       "bytes": 1183559,
@@ -562,6 +556,12 @@
       "bytes": 5869955,
       "pages": 40,
       "shards": 2
+    },
+    {
+      "seconds": 1.1126535420044092,
+      "bytes": 5839811,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -416,3 +416,4 @@
 {"step": 926600, "loss": 0.504806399345398, "lr": 3e-05, "grad_norm": 0.3150711242582432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135627853641, "tokens_per_second": 152076.48591215358, "data_wait_s": 1.6559002688154578e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 926700, "loss": 0.5237295031547546, "lr": 3e-05, "grad_norm": 0.33451088024721903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135642531350, "tokens_per_second": 152937.22285901834, "data_wait_s": 1.7312995623797178e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 926800, "loss": 0.5488107800483704, "lr": 3e-05, "grad_norm": 0.35696342018321464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135657183960, "tokens_per_second": 152605.09608179412, "data_wait_s": 1.4655000995844603e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 926900, "loss": 0.5539249181747437, "lr": 3e-05, "grad_norm": 0.3245176664790011, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135671791690, "tokens_per_second": 153431.22632042432, "data_wait_s": 1.327700010733679e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5604bfd6d051f6bb9751737e46f9fff12611264c8e306544706a8909c520751c
-size 891753
+oid sha256:90792b8b5cb698bbd2bd5c18cc950fc00706a5c01c457eefe24bfe24ba2c7c26
+size 893871