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": 78274.90661615318,
-      "required_pages_per_second_upper_estimate": 0.005415239662511566
+      "required_token_positions_per_second": 76041.6329384492,
+      "required_pages_per_second_upper_estimate": 0.005260736607579144
     },
     {
       "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": 9784.363327019148,
-      "required_pages_per_second_upper_estimate": 0.004096508604282563
+      "required_token_positions_per_second": 9505.20411730615,
+      "required_pages_per_second_upper_estimate": 0.003979630472682925
     },
     {
       "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": 19568.726654038295,
-      "required_pages_per_second_upper_estimate": 0.004075649700010642
+      "required_token_positions_per_second": 19010.4082346123,
+      "required_pages_per_second_upper_estimate": 0.003959366697091063
     },
     {
       "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": 29353.089981057445,
-      "required_pages_per_second_upper_estimate": 0.0017596729247911303
+      "required_token_positions_per_second": 28515.61235191845,
+      "required_pages_per_second_upper_estimate": 0.001709467419678545
     },
     {
       "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": 29353.089981057445,
-      "required_pages_per_second_upper_estimate": 0.0022820736354292183
+      "required_token_positions_per_second": 28515.61235191845,
+      "required_pages_per_second_upper_estimate": 0.0022169634334384494
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 29353.089981057445,
-      "required_pages_per_second_upper_estimate": 0.3765099213845056
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28515.61235191845,
+      "required_pages_per_second_upper_estimate": 0.3657676575713299
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.04762147552307589,
-    "producer_p95_seconds": 0.1201721610000277,
-    "gpu_update_seconds": 0.9419110566499882,
-    "producer_batches_per_second": 20.99892934891173,
-    "gpu_batches_per_second": 1.0616713679491263,
-    "producer_headroom": 19.7791237315519,
+    "producer_batch_seconds": 0.0673477296718783,
+    "producer_p95_seconds": 0.12105643900008545,
+    "gpu_update_seconds": 0.9695741286839284,
+    "producer_batches_per_second": 14.848310475676803,
+    "gpu_batches_per_second": 1.0313806550896432,
+    "producer_headroom": 14.396537691882784,
     "sustainable": true,
-    "buffer_seconds": 60.282307625599245,
+    "buffer_seconds": 60.113595978403566,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 18,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 60,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 397195.83696778223,
-    "upload_bytes_per_second": 104461.87231221441,
-    "upload_sustainable": false,
+    "generation_bytes_per_second": 993438.9054075802,
+    "upload_bytes_per_second": 1933301.9964728944,
+    "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.22898036742306,
+    "estimated_full_shards_per_hour": 106.58443151316906,
     "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": [
@@ -190,6 +190,12 @@
       "bytes": 145464,
       "pages": 4,
       "shards": 1
+    },
+    {
+      "seconds": 1.1176107269999989,
+      "bytes": 5541224,
+      "pages": 18,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -1 +1,2 @@
 {"step": 557200, "loss": 0.5685006380081177, "lr": 0.0003, "grad_norm": 0.20223924821874198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81556112305, "tokens_per_second": 148940.87266347083, "data_wait_s": 1.2630000014723919e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 557300, "loss": 0.5449313521385193, "lr": 0.0003, "grad_norm": 0.16618840335759535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81570754119, "tokens_per_second": 152862.13565577025, "data_wait_s": 1.6536999964955612e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e1aa75330a74cd1308fa2b0cec3db103585aaf8b4e08867a2a260a365f3ac548
-size 2283
+oid sha256:d473fe5fad0d9e470dc16bf738fd7e63ad9f7d69e6438f7441f77f1c9247e892
+size 4401