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": 76794.85183442647,
-      "required_pages_per_second_upper_estimate": 0.005312846038511478
+      "required_token_positions_per_second": 77445.12249503432,
+      "required_pages_per_second_upper_estimate": 0.005357833271648138
     },
     {
       "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": 9599.35647930331,
-      "required_pages_per_second_upper_estimate": 0.0040190501005262416
+      "required_token_positions_per_second": 9680.64031187929,
+      "required_pages_per_second_upper_estimate": 0.004053081943826363
     },
     {
       "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": 19198.71295860662,
-      "required_pages_per_second_upper_estimate": 0.003998585605169564
+      "required_token_positions_per_second": 19361.28062375858,
+      "required_pages_per_second_upper_estimate": 0.004032444162623086
     },
     {
       "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": 28798.069437909926,
-      "required_pages_per_second_upper_estimate": 0.001726400290696736
+      "required_token_positions_per_second": 29041.92093563787,
+      "required_pages_per_second_upper_estimate": 0.0017410188156458484
     },
     {
       "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": 28798.069437909926,
-      "required_pages_per_second_upper_estimate": 0.0022389232294768658
+      "required_token_positions_per_second": 29041.92093563787,
+      "required_pages_per_second_upper_estimate": 0.0022578816108358296
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28798.069437909926,
-      "required_pages_per_second_upper_estimate": 0.36939071379163846
+      "required_token_positions_per_second": 29041.92093563787,
+      "required_pages_per_second_upper_estimate": 0.37251857897715357
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0990487014998962,
-    "producer_p95_seconds": 0.29144657899814774,
-    "gpu_update_seconds": 0.9600643563836967,
-    "producer_batches_per_second": 10.096043510485071,
-    "gpu_batches_per_second": 1.041596840202182,
-    "producer_headroom": 9.692851514915649,
+    "producer_batch_seconds": 0.3554747217030183,
+    "producer_p95_seconds": 0.297676310998213,
+    "gpu_update_seconds": 0.9520031426734117,
+    "producer_batches_per_second": 2.813139553803353,
+    "gpu_batches_per_second": 1.0504167005077354,
+    "producer_headroom": 2.678117695999671,
     "sustainable": true,
-    "buffer_seconds": 60.48405445217289,
+    "buffer_seconds": 60.92820113109835,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 409549.3652094244,
-    "upload_bytes_per_second": 5446493.523182744,
+    "generation_bytes_per_second": 355460.179973465,
+    "upload_bytes_per_second": 5514271.160912827,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.87975995266008,
+    "estimated_full_shards_per_hour": 76.27347993281329,
     "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": 1.3234173230011947,
-      "bytes": 2631414,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.286528674998408,
       "bytes": 6199355,
@@ -562,6 +556,12 @@
       "bytes": 1199966,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 1.69540837599925,
+      "bytes": 9973657,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -295,3 +295,4 @@
 {"step": 674500, "loss": 0.5711096525192261, "lr": 0.0001, "grad_norm": 0.19743078376877174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98728999786, "tokens_per_second": 150786.6746819291, "data_wait_s": 1.1530999472597614e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 674600, "loss": 0.5266985297203064, "lr": 0.0001, "grad_norm": 0.19269618392785934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98743645907, "tokens_per_second": 152206.86829658982, "data_wait_s": 9.821000276133418e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 674700, "loss": 0.5378531217575073, "lr": 0.0001, "grad_norm": 0.19060410120226906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98758370154, "tokens_per_second": 153270.14406059575, "data_wait_s": 1.1246000212850049e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 674800, "loss": 0.5367597937583923, "lr": 0.0001, "grad_norm": 0.1808463882216845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98773094831, "tokens_per_second": 152515.78546607023, "data_wait_s": 1.2541997421067208e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:965a20e0c273e7223fa3fcfe0590a7fdcb2349fdcecc07633530e842bbec732b
-size 633126
+oid sha256:d497de3761438d63f7bd010dcd881f1660ede3ca82e309c71efeb61aa5333f28
+size 635244