Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74850.46237933666,
-      "required_pages_per_second_upper_estimate": 0.006487848372278385
+      "required_token_positions_per_second": 75145.61086443164,
+      "required_pages_per_second_upper_estimate": 0.006513431094919421
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9356.307797417083,
-      "required_pages_per_second_upper_estimate": 0.003732371069657365
+      "required_token_positions_per_second": 9393.201358053955,
+      "required_pages_per_second_upper_estimate": 0.003747088462603301
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18712.615594834166,
-      "required_pages_per_second_upper_estimate": 0.004328834287078458
+      "required_token_positions_per_second": 18786.40271610791,
+      "required_pages_per_second_upper_estimate": 0.0043459036389761575
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28068.92339225125,
-      "required_pages_per_second_upper_estimate": 0.0017092501683283981
+      "required_token_positions_per_second": 28179.604074161867,
+      "required_pages_per_second_upper_estimate": 0.0017159900411600954
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28068.92339225125,
-      "required_pages_per_second_upper_estimate": 0.0021608234224884566
+      "required_token_positions_per_second": 28179.604074161867,
+      "required_pages_per_second_upper_estimate": 0.0021693439277656727
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28068.92339225125,
-      "required_pages_per_second_upper_estimate": 0.34852208788819117
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28179.604074161867,
+      "required_pages_per_second_upper_estimate": 0.3498963715331074
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.14964061438263343,
-    "producer_p95_seconds": 0.40974565599753987,
-    "gpu_update_seconds": 0.9850039352643126,
-    "producer_batches_per_second": 6.682677721724558,
-    "gpu_batches_per_second": 1.015224370379458,
-    "producer_headroom": 6.582463854001841,
+    "producer_batch_seconds": 0.27247613208567145,
+    "producer_p95_seconds": 0.43742603499413235,
+    "gpu_update_seconds": 0.9811351475072959,
+    "producer_batches_per_second": 3.670046225133517,
+    "gpu_batches_per_second": 1.0192275779138407,
+    "producer_headroom": 3.6008113444549674,
     "sustainable": true,
-    "buffer_seconds": 60.08524005112307,
+    "buffer_seconds": 60.83037914545235,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 171,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 631105.9934122637,
-    "upload_bytes_per_second": 5791854.595441407,
+    "generation_bytes_per_second": 602009.0039009083,
+    "upload_bytes_per_second": 5755809.57296835,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.71032739532438,
+    "estimated_full_shards_per_hour": 64.58855909238069,
     "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": 1.0351565999953891,
-      "bytes": 6216009,
-      "pages": 41,
-      "shards": 2
-    },
     {
       "seconds": 1.4164216759963892,
       "bytes": 2450413,
@@ -562,6 +556,12 @@
       "bytes": 6056911,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 0.8230368889999227,
+      "bytes": 2438177,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -406,3 +406,4 @@
 {"step": 155600, "loss": 0.6896321140229702, "lr": 0.0003, "grad_norm": 0.16904787321003917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22772124076, "tokens_per_second": 149032.15146045198, "data_wait_s": 1.2080003216397017e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 155700, "loss": 0.5687103159725666, "lr": 0.0003, "grad_norm": 0.14221636869850315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22786637208, "tokens_per_second": 147880.678066207, "data_wait_s": 1.4651996025349945e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 155800, "loss": 0.5813593603670597, "lr": 0.0003, "grad_norm": 0.1494349498685959, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22801290176, "tokens_per_second": 148898.42387125635, "data_wait_s": 1.0771000233944505e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 155900, "loss": 0.5951719842851162, "lr": 0.0003, "grad_norm": 0.15011265849159988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22815968278, "tokens_per_second": 148507.44070253478, "data_wait_s": 1.1504002031870186e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a07b23f4debd2faaf27c4af8f25a670609db97cebb8434f076ca803e537e7c72
-size 870573
+oid sha256:cd348a7b4a4ac3f02e606a697edf1a2ccb805a93f8b1533078e60babd8b5f71c
+size 872691