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": 74938.81896406057,
-      "required_pages_per_second_upper_estimate": 0.006495506897104536
+      "required_token_positions_per_second": 73461.32813647029,
+      "required_pages_per_second_upper_estimate": 0.006367441736835278
     },
     {
       "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": 9367.352370507571,
-      "required_pages_per_second_upper_estimate": 0.0037367769149942442
+      "required_token_positions_per_second": 9182.666017058786,
+      "required_pages_per_second_upper_estimate": 0.003663102767296468
     },
     {
       "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": 18734.704741015143,
-      "required_pages_per_second_upper_estimate": 0.004333944222291732
+      "required_token_positions_per_second": 18365.33203411757,
+      "required_pages_per_second_upper_estimate": 0.004248496347288593
     },
     {
       "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": 28102.05711152271,
-      "required_pages_per_second_upper_estimate": 0.00171126783799284
+      "required_token_positions_per_second": 27547.998051176357,
+      "required_pages_per_second_upper_estimate": 0.0016775285481409779
     },
     {
       "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": 28102.05711152271,
-      "required_pages_per_second_upper_estimate": 0.0021633741479179787
+      "required_token_positions_per_second": 27547.998051176357,
+      "required_pages_per_second_upper_estimate": 0.002120721147718874
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28102.05711152271,
-      "required_pages_per_second_upper_estimate": 0.34893349779011773
+      "required_token_positions_per_second": 27547.998051176357,
+      "required_pages_per_second_upper_estimate": 0.34205393857700633
     }
   ],
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.15063954439852978,
-    "producer_p95_seconds": 0.39601011400372954,
-    "gpu_update_seconds": 0.9838425667658139,
-    "producer_batches_per_second": 6.638363146893319,
-    "gpu_batches_per_second": 1.0164227832582,
-    "producer_headroom": 6.531104237563109,
+    "producer_batch_seconds": 0.10039969090655632,
+    "producer_p95_seconds": 0.3884768560019438,
+    "gpu_update_seconds": 1.003630098587849,
+    "producer_batches_per_second": 9.960190026189592,
+    "gpu_batches_per_second": 0.9963830313648856,
+    "producer_headroom": 9.99634649793837,
     "sustainable": true,
-    "buffer_seconds": 60.01439657271465,
+    "buffer_seconds": 60.217805915270944,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 370807.939579464,
-    "upload_bytes_per_second": 3793650.304880353,
+    "generation_bytes_per_second": 381983.93150267703,
+    "upload_bytes_per_second": 3772732.811162595,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.56675186670246,
+    "estimated_full_shards_per_hour": 81.96485956964715,
     "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": 0.9863674269981857,
-      "bytes": 7525026,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.7792704010025773,
       "bytes": 5012728,
@@ -562,6 +556,12 @@
       "bytes": 2689085,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1650014270053362,
+      "bytes": 5886860,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -355,3 +355,4 @@
 {"step": 35500, "loss": 0.6256066113710403, "lr": 0.0003, "grad_norm": 0.1355154730981297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5194677629, "tokens_per_second": 149382.83431856887, "data_wait_s": 9.512004908174276e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 35600, "loss": 0.5915225241333246, "lr": 0.0003, "grad_norm": 0.1300828179687159, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5209305680, "tokens_per_second": 148522.84898063625, "data_wait_s": 1.2244003301020712e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 35700, "loss": 0.6167456582188606, "lr": 0.0003, "grad_norm": 0.1455468630375116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5223942532, "tokens_per_second": 149018.186610883, "data_wait_s": 1.1085998266935349e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 35800, "loss": 0.6350256651639938, "lr": 0.0003, "grad_norm": 0.12678470938493638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5238589021, "tokens_per_second": 148927.44061422633, "data_wait_s": 1.4830999134574085e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f56b586e89fee79159ec0e1a8b537727e2bcd479da26abea665569f467dee764
-size 755998
+oid sha256:63344a9ae3fcb336374a4794d702853d9d50c18696104d1ddde55ab855293ee5
+size 758116