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": 74601.81223629815,
-      "required_pages_per_second_upper_estimate": 0.0064662960080778325
+      "required_token_positions_per_second": 74990.32336289072,
+      "required_pages_per_second_upper_estimate": 0.006499971167858429
     },
     {
       "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": 9325.226529537269,
-      "required_pages_per_second_upper_estimate": 0.003719972287193741
+      "required_token_positions_per_second": 9373.79042036134,
+      "required_pages_per_second_upper_estimate": 0.003739345149338336
     },
     {
       "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": 18650.453059074538,
-      "required_pages_per_second_upper_estimate": 0.004314454078453732
+      "required_token_positions_per_second": 18747.58084072268,
+      "required_pages_per_second_upper_estimate": 0.004336922881347456
     },
     {
       "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": 27975.679588611805,
-      "required_pages_per_second_upper_estimate": 0.0017035721098991854
+      "required_token_positions_per_second": 28121.37126108402,
+      "required_pages_per_second_upper_estimate": 0.0017124439683676126
     },
     {
       "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": 27975.679588611805,
-      "required_pages_per_second_upper_estimate": 0.0021536452563689235
+      "required_token_positions_per_second": 28121.37126108402,
+      "required_pages_per_second_upper_estimate": 0.002164861004616196
     },
     {
       "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": 27975.679588611805,
-      "required_pages_per_second_upper_estimate": 0.34736431191392536
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28121.37126108402,
+      "required_pages_per_second_upper_estimate": 0.3491733148873688
     }
   ],
   "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": 0.4314636074139173,
+    "producer_batch_seconds": 0.07765354547652237,
     "producer_p95_seconds": 0.25262037200081977,
-    "gpu_update_seconds": 0.98828698378626,
-    "producer_batches_per_second": 2.3176925766549457,
-    "gpu_batches_per_second": 1.01185183697236,
-    "producer_headroom": 2.2905454059261214,
+    "gpu_update_seconds": 0.9831668499843357,
+    "producer_batches_per_second": 12.877712071786062,
+    "gpu_batches_per_second": 1.0171213563760133,
+    "producer_headroom": 12.660939612623155,
     "sustainable": true,
-    "buffer_seconds": 60.28550601096186,
+    "buffer_seconds": 60.95634469902881,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340173.5828001034,
-    "upload_bytes_per_second": 5053820.198197049,
+    "generation_bytes_per_second": 339960.8487311269,
+    "upload_bytes_per_second": 5134583.5224400675,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.99333203317954,
+    "estimated_full_shards_per_hour": 72.94768425417286,
     "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": 0.9772154700003739,
-      "bytes": 2553165,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.3266675979994034,
       "bytes": 5992849,
@@ -562,6 +556,12 @@
       "bytes": 1237044,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.392788182998629,
+      "bytes": 11322146,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -166,3 +166,4 @@
 {"step": 171600, "loss": 0.5917437970638275, "lr": 0.0003, "grad_norm": 0.1492916433478241, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25114568192, "tokens_per_second": 148368.72933766892, "data_wait_s": 1.4881999959470704e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 171700, "loss": 0.6072972640395164, "lr": 0.0003, "grad_norm": 0.15555167915432963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25129335930, "tokens_per_second": 144262.320093169, "data_wait_s": 1.617499947315082e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 171800, "loss": 0.589970288798213, "lr": 0.0003, "grad_norm": 0.17102003820212547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25144044313, "tokens_per_second": 149047.31067176312, "data_wait_s": 1.2485001207096502e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 171900, "loss": 0.6244958490133286, "lr": 0.0003, "grad_norm": 0.14701528131234995, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25158621255, "tokens_per_second": 148934.09620441706, "data_wait_s": 1.2003001756966114e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9a6f1c411b8e7a850bb4f3584aeb29ed2971491f0df7dd99db7117baa44a2e6c
-size 358338
+oid sha256:120ab76d7ecb761444115b29a1b0436b706b6a94215909ffc7d0b2a1a39621b8
+size 360456