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": 75692.56212146599,
-      "required_pages_per_second_upper_estimate": 0.006560839443643871
+      "required_token_positions_per_second": 75256.60734365601,
+      "required_pages_per_second_upper_estimate": 0.006523051988420585
     },
     {
       "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": 9461.570265183249,
-      "required_pages_per_second_upper_estimate": 0.003774361841863431
+      "required_token_positions_per_second": 9407.075917957001,
+      "required_pages_per_second_upper_estimate": 0.003752623231991783
     },
     {
       "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": 18923.140530366498,
-      "required_pages_per_second_upper_estimate": 0.004377535525801543
+      "required_token_positions_per_second": 18814.151835914003,
+      "required_pages_per_second_upper_estimate": 0.0043523229094754685
     },
     {
       "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": 28384.710795549745,
-      "required_pages_per_second_upper_estimate": 0.0017284799643808163
+      "required_token_positions_per_second": 28221.227753871004,
+      "required_pages_per_second_upper_estimate": 0.0017185247048718119
     },
     {
       "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": 28384.710795549745,
-      "required_pages_per_second_upper_estimate": 0.0021851336109498555
+      "required_token_positions_per_second": 28221.227753871004,
+      "required_pages_per_second_upper_estimate": 0.002172548233851402
     },
     {
       "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": 28384.710795549745,
-      "required_pages_per_second_upper_estimate": 0.35244311056470623
+      "required_token_positions_per_second": 28221.227753871004,
+      "required_pages_per_second_upper_estimate": 0.3504131983296001
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.5476751618515436,
-    "producer_p95_seconds": 0.3958749640005408,
-    "gpu_update_seconds": 0.974045506369392,
-    "producer_batches_per_second": 1.8258998575346501,
-    "gpu_batches_per_second": 1.0266460791214462,
-    "producer_headroom": 1.7785095513121392,
+    "producer_batch_seconds": 0.12017607360184002,
+    "producer_p95_seconds": 0.39018070100428304,
+    "gpu_update_seconds": 0.979688064641611,
+    "producer_batches_per_second": 8.321123914508462,
+    "gpu_batches_per_second": 1.0207330640144314,
+    "producer_headroom": 8.152105783447821,
     "sustainable": true,
-    "buffer_seconds": 60.390821394902304,
+    "buffer_seconds": 60.74066000777988,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386296.1007248937,
-    "upload_bytes_per_second": 5676285.960294976,
+    "generation_bytes_per_second": 387379.8646465509,
+    "upload_bytes_per_second": 5773821.700815944,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.8901507025729,
+    "estimated_full_shards_per_hour": 83.12270121142764,
     "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": 1.3384703369993076,
-      "bytes": 7381158,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 0.9186418360004609,
       "bytes": 6519513,
@@ -562,6 +556,12 @@
       "bytes": 5559059,
       "pages": 23,
       "shards": 2
+    },
+    {
+      "seconds": 1.564358691997768,
+      "bytes": 15793806,
+      "pages": 43,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -361,3 +361,4 @@
 {"step": 231100, "loss": 0.5924375485628843, "lr": 0.0003, "grad_norm": 0.15358562835226264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33826888084, "tokens_per_second": 148428.65143981823, "data_wait_s": 1.4299002941697836e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 231200, "loss": 0.6378389857709408, "lr": 0.0003, "grad_norm": 0.15674925291480424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33841347076, "tokens_per_second": 146559.54507288084, "data_wait_s": 1.1919000826310366e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 231300, "loss": 0.5928157931193709, "lr": 0.0003, "grad_norm": 0.13617362397365723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33855910977, "tokens_per_second": 147701.33501886614, "data_wait_s": 1.848700048867613e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 231400, "loss": 0.582163080573082, "lr": 0.0003, "grad_norm": 0.13887501363023447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33870486465, "tokens_per_second": 148673.70855027012, "data_wait_s": 1.2397998943924904e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3a28eb6275cb10e37cbd5cde502f60747b2d4ff049d920a0ee3de0ec047bd689
-size 774480
+oid sha256:28b93d777f7fa75706f706d17f887d9616317407db9ca17bd7f3b33e1a37d230
+size 776598