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": 75743.64578770011,
-      "required_pages_per_second_upper_estimate": 0.006565267246362676
+      "required_token_positions_per_second": 76001.74627129106,
+      "required_pages_per_second_upper_estimate": 0.006587638742130645
     },
     {
       "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": 9467.955723462514,
-      "required_pages_per_second_upper_estimate": 0.003776909096642139
+      "required_token_positions_per_second": 9500.218283911383,
+      "required_pages_per_second_upper_estimate": 0.0037897791143734574
     },
     {
       "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": 18935.911446925027,
-      "required_pages_per_second_upper_estimate": 0.004380489852586904
+      "required_token_positions_per_second": 19000.436567822766,
+      "required_pages_per_second_upper_estimate": 0.004395416603703254
     },
     {
       "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": 28403.86717038754,
-      "required_pages_per_second_upper_estimate": 0.0017296464871027067
+      "required_token_positions_per_second": 28500.654851734147,
+      "required_pages_per_second_upper_estimate": 0.0017355403490909964
     },
     {
       "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": 28403.86717038754,
-      "required_pages_per_second_upper_estimate": 0.0021866083217131085
+      "required_token_positions_per_second": 28500.654851734147,
+      "required_pages_per_second_upper_estimate": 0.00219405930534862
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28403.86717038754,
-      "required_pages_per_second_upper_estimate": 0.3526809686279293
+      "required_token_positions_per_second": 28500.654851734147,
+      "required_pages_per_second_upper_estimate": 0.3538827477027223
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08428846096077791,
-    "producer_p95_seconds": 0.3381703790000756,
-    "gpu_update_seconds": 0.97338858241192,
-    "producer_batches_per_second": 11.864020158884282,
-    "gpu_batches_per_second": 1.0273389456882067,
-    "producer_headroom": 11.548301764162813,
+    "producer_batch_seconds": 0.27858332232028715,
+    "producer_p95_seconds": 0.3454707429991686,
+    "gpu_update_seconds": 0.970082973315181,
+    "producer_batches_per_second": 3.5895903303583276,
+    "gpu_batches_per_second": 1.0308396575424676,
+    "producer_headroom": 3.4822004606574293,
     "sustainable": true,
-    "buffer_seconds": 60.350092109539034,
+    "buffer_seconds": 60.145144345541226,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 454964.39878162264,
+    "generation_bytes_per_second": 447904.08661895216,
     "upload_bytes_per_second": 3998840.12879947,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.62476894938001,
+    "estimated_full_shards_per_hour": 96.10979031492637,
     "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": [

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -253,3 +253,4 @@
 {"step": 65300, "loss": 0.6247148029506207, "lr": 0.0003, "grad_norm": 0.12495317407980254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9556871900, "tokens_per_second": 150364.37765044507, "data_wait_s": 9.878000128082931e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 65400, "loss": 0.5901113897562027, "lr": 0.0003, "grad_norm": 0.12827208096657697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9571518480, "tokens_per_second": 148836.67122486653, "data_wait_s": 1.136900027631782e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 65500, "loss": 0.6445927396416664, "lr": 0.0003, "grad_norm": 0.1270078281420115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9586195991, "tokens_per_second": 149760.59558858935, "data_wait_s": 1.6928999684751034e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 65600, "loss": 0.6546289436519146, "lr": 0.0003, "grad_norm": 0.15052764836424393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9600814721, "tokens_per_second": 148990.2884754614, "data_wait_s": 1.0017000022344291e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:01d88541d32e9a22b8db2f4a3f46b50432302d8756548eb14bec6c28fe792124
-size 544170
+oid sha256:ca11f2c87d9cef5adcdb9133497171ae0944956ad1745a114e7eb46918c13d62
+size 546288