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": 75803.89389061033,
-      "required_pages_per_second_upper_estimate": 0.006570489399225513
+      "required_token_positions_per_second": 75514.77040157851,
+      "required_pages_per_second_upper_estimate": 0.0065454289079730195
     },
     {
       "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": 9475.486736326291,
-      "required_pages_per_second_upper_estimate": 0.003779913330272176
+      "required_token_positions_per_second": 9439.346300197314,
+      "required_pages_per_second_upper_estimate": 0.0037654963699526546
     },
     {
       "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": 18950.973472652582,
-      "required_pages_per_second_upper_estimate": 0.004383974187156375
+      "required_token_positions_per_second": 18878.69260039463,
+      "required_pages_per_second_upper_estimate": 0.0043672532794594016
     },
     {
       "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": 28426.46020897887,
-      "required_pages_per_second_upper_estimate": 0.0017310222846164064
+      "required_token_positions_per_second": 28318.038900591942,
+      "required_pages_per_second_upper_estimate": 0.001724419995778285
     },
     {
       "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": 28426.46020897887,
-      "required_pages_per_second_upper_estimate": 0.0021883475963654026
+      "required_token_positions_per_second": 28318.038900591942,
+      "required_pages_per_second_upper_estimate": 0.0021800010239163862
     },
     {
       "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": 28426.46020897887,
-      "required_pages_per_second_upper_estimate": 0.35296149855319753
+      "required_token_positions_per_second": 28318.038900591942,
+      "required_pages_per_second_upper_estimate": 0.3516152687658088
     }
   ],
   "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": 0.09691960866422278,
-    "producer_p95_seconds": 0.30038088600122137,
-    "gpu_update_seconds": 0.9726149438496396,
-    "producer_batches_per_second": 10.317829526783296,
-    "gpu_batches_per_second": 1.0281561128826269,
-    "producer_headroom": 10.03527518584249,
+    "producer_batch_seconds": 0.3315773975782008,
+    "producer_p95_seconds": 0.3585726179990161,
+    "gpu_update_seconds": 0.9763387958133662,
+    "producer_batches_per_second": 3.0158871120404256,
+    "gpu_batches_per_second": 1.0242346245873821,
+    "producer_headroom": 2.9445275912785998,
     "sustainable": true,
-    "buffer_seconds": 60.30212651867766,
+    "buffer_seconds": 60.53300534042871,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 391174.08952095744,
+    "generation_bytes_per_second": 390379.22666107246,
     "upload_bytes_per_second": 5702512.751565578,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.93685354444068,
+    "estimated_full_shards_per_hour": 83.76629447816973,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -237,3 +237,4 @@
 {"step": 23700, "loss": 0.6171144619584084, "lr": 0.0003, "grad_norm": 0.13719628169519021, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3467715552, "tokens_per_second": 150272.31114418886, "data_wait_s": 1.2953998520970345e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 23800, "loss": 0.6605587769299746, "lr": 0.0003, "grad_norm": 0.13749161686068306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3482246711, "tokens_per_second": 147306.87022921073, "data_wait_s": 1.0925999958999455e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 23900, "loss": 0.7071166168898344, "lr": 0.0003, "grad_norm": 0.14389220587248958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3496953173, "tokens_per_second": 149814.61366868, "data_wait_s": 1.301900192629546e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 24000, "loss": 0.7391644399613142, "lr": 0.0003, "grad_norm": 0.14073769029739688, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3511650931, "tokens_per_second": 149098.6586843954, "data_wait_s": 1.089999932446517e-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 sha256:8e75ef10f4fafaf21bb5761ebf8a78077a7e5c518ae3d8ab9c2dbdbd7666e341
-size 503725
+oid sha256:651290e665f6ab9e3937c63becb7381700f4910653838ef19449ea11798b37d6
+size 505843