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": 74991.01741599824,
-      "required_pages_per_second_upper_estimate": 0.006500031326622727
+      "required_token_positions_per_second": 75340.44722879169,
+      "required_pages_per_second_upper_estimate": 0.006530319017173899
     },
     {
       "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": 9373.87717699978,
-      "required_pages_per_second_upper_estimate": 0.003739379757858537
+      "required_token_positions_per_second": 9417.55590359896,
+      "required_pages_per_second_upper_estimate": 0.003756803854954109
     },
     {
       "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": 18747.75435399956,
-      "required_pages_per_second_upper_estimate": 0.0043369630205903325
+      "required_token_positions_per_second": 18835.11180719792,
+      "required_pages_per_second_upper_estimate": 0.004357171629949108
     },
     {
       "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": 28121.63153099934,
-      "required_pages_per_second_upper_estimate": 0.001712459817439925
+      "required_token_positions_per_second": 28252.667710796883,
+      "required_pages_per_second_upper_estimate": 0.0017204392332958922
     },
     {
       "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": 28121.63153099934,
-      "required_pages_per_second_upper_estimate": 0.0021648810409147508
+      "required_token_positions_per_second": 28252.667710796883,
+      "required_pages_per_second_upper_estimate": 0.002174968568766927
     },
     {
       "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": 28121.63153099934,
-      "required_pages_per_second_upper_estimate": 0.3491765465686497
+      "required_token_positions_per_second": 28252.667710796883,
+      "required_pages_per_second_upper_estimate": 0.35080357737185247
     }
   ],
   "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": 0.1203048181563986,
-    "producer_p95_seconds": 0.4188161720012431,
-    "gpu_update_seconds": 0.9831577506277599,
-    "producer_batches_per_second": 8.312219039307143,
-    "gpu_batches_per_second": 1.0171307700737608,
-    "producer_headroom": 8.17222257341045,
+    "producer_batch_seconds": 0.10456494309391928,
+    "producer_p95_seconds": 0.4139775340008782,
+    "gpu_update_seconds": 0.9785978542987002,
+    "producer_batches_per_second": 9.563434650385732,
+    "gpu_batches_per_second": 1.0218702152342622,
+    "producer_headroom": 9.358756628593317,
     "sustainable": true,
-    "buffer_seconds": 60.95578053892111,
+    "buffer_seconds": 60.67306696651941,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 292872.8690905846,
-    "upload_bytes_per_second": 5846277.8884541895,
+    "generation_bytes_per_second": 303911.47314950684,
+    "upload_bytes_per_second": 5794773.637208084,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.84369997537759,
+    "estimated_full_shards_per_hour": 65.21232744087128,
     "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": 1.181266768999194,
-      "bytes": 6999855,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9345153110007232,
       "bytes": 6342336,
@@ -562,6 +556,12 @@
       "bytes": 11558434,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 0.8286585169989849,
+      "bytes": 1181916,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -386,3 +386,4 @@
 {"step": 233600, "loss": 0.6095574572682381, "lr": 0.0003, "grad_norm": 0.17349696892607225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34192695865, "tokens_per_second": 146619.6809009955, "data_wait_s": 1.079399953596294e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 233700, "loss": 0.5399864874780178, "lr": 0.0003, "grad_norm": 0.1503727513628934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34207228558, "tokens_per_second": 147259.34759513917, "data_wait_s": 8.631999662611634e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 233800, "loss": 0.5587761066854, "lr": 0.0003, "grad_norm": 0.1490171192827732, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34221764765, "tokens_per_second": 147533.8362568862, "data_wait_s": 1.1696000001393259e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 233900, "loss": 0.5574611648917198, "lr": 0.0003, "grad_norm": 0.14534793742231727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34236369768, "tokens_per_second": 148103.06027596357, "data_wait_s": 1.2780001270584762e-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 sha256:7be56aa3857e47034e850fa22b8c6500ba5bdd1072c51e452c46e47f7ac44aea
-size 827430
+oid sha256:1b43696ebc8048b73af2d20971ce21137de6ec0674751c05f1383da18ad6ac3f
+size 829548