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": 76608.48295631792,
-      "required_pages_per_second_upper_estimate": 0.005299952607088553
+      "required_token_positions_per_second": 76684.42579237987,
+      "required_pages_per_second_upper_estimate": 0.005305206508698976
     },
     {
       "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": 9576.06036953974,
-      "required_pages_per_second_upper_estimate": 0.0040092965058463265
+      "required_token_positions_per_second": 9585.553224047484,
+      "required_pages_per_second_upper_estimate": 0.0040132709657953744
     },
     {
       "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": 19152.12073907948,
-      "required_pages_per_second_upper_estimate": 0.003988881674561517
+      "required_token_positions_per_second": 19171.106448094968,
+      "required_pages_per_second_upper_estimate": 0.0039928358970626274
     },
     {
       "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": 28728.18110861922,
-      "required_pages_per_second_upper_estimate": 0.0017222105920690581
+      "required_token_positions_per_second": 28756.65967214245,
+      "required_pages_per_second_upper_estimate": 0.001723917838468027
     },
     {
       "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": 28728.18110861922,
-      "required_pages_per_second_upper_estimate": 0.0022334897192807864
+      "required_token_positions_per_second": 28756.65967214245,
+      "required_pages_per_second_upper_estimate": 0.0022357038023307547
     },
     {
       "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": 28728.18110861922,
-      "required_pages_per_second_upper_estimate": 0.3684942613437388
+      "required_token_positions_per_second": 28756.65967214245,
+      "required_pages_per_second_upper_estimate": 0.36885955377871565
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08617622139058767,
-    "producer_p95_seconds": 0.22807512899817084,
-    "gpu_update_seconds": 0.962399947823528,
-    "producer_batches_per_second": 11.604129118954638,
-    "gpu_batches_per_second": 1.0390690505142948,
-    "producer_headroom": 11.167813258619427,
+    "producer_batch_seconds": 0.06371678670328151,
+    "producer_p95_seconds": 0.21765600300204824,
+    "gpu_update_seconds": 0.9614468549274363,
+    "producer_batches_per_second": 15.69445120729069,
+    "gpu_batches_per_second": 1.0400990911509855,
+    "producer_headroom": 15.089380753061741,
     "sustainable": true,
-    "buffer_seconds": 60.631196712882264,
+    "buffer_seconds": 60.571151860428486,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334302.0562461318,
-    "upload_bytes_per_second": 4757192.28809234,
+    "generation_bytes_per_second": 332749.12434625643,
+    "upload_bytes_per_second": 4747579.128254849,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.73343911684003,
+    "estimated_full_shards_per_hour": 71.40021608153124,
     "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": 1.0955427600001713,
-      "bytes": 6199869,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.6065525880003406,
       "bytes": 12580162,
@@ -562,6 +556,12 @@
       "bytes": 11619395,
       "pages": 28,
       "shards": 4
+    },
+    {
+      "seconds": 1.2608952039990982,
+      "bytes": 6128515,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -180,3 +180,4 @@
 {"step": 415100, "loss": 0.5824700593948364, "lr": 0.0003, "grad_norm": 0.15353316396598302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60759930315, "tokens_per_second": 117552.91918434553, "data_wait_s": 1.4602999726776034e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 415200, "loss": 0.5866156816482544, "lr": 0.0003, "grad_norm": 0.16299583387780173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60774386259, "tokens_per_second": 149488.56135730998, "data_wait_s": 1.4161996659822762e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 415300, "loss": 0.586100697517395, "lr": 0.0003, "grad_norm": 0.18045619156312825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60788927038, "tokens_per_second": 150968.94529780687, "data_wait_s": 1.1275998986093327e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 415400, "loss": 0.587268054485321, "lr": 0.0003, "grad_norm": 0.16183639088747342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60803444346, "tokens_per_second": 150496.9862530765, "data_wait_s": 1.4606001059291884e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:706f846f2e02d8f7fdcbd21982619e253a4a944bb84b88bc39068ffc8d27bbe3
-size 388773
+oid sha256:5f014691a4255bf4db0de0470326f4227cff86c6efdf0978a8086b0b41247358
+size 390891