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": 76018.80090035984,
-      "required_pages_per_second_upper_estimate": 0.006589116994153764
+      "required_token_positions_per_second": 75665.30160116269,
+      "required_pages_per_second_upper_estimate": 0.006558476570835138
     },
     {
       "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": 9502.35011254498,
-      "required_pages_per_second_upper_estimate": 0.003790629532689078
+      "required_token_positions_per_second": 9458.162700145336,
+      "required_pages_per_second_upper_estimate": 0.00377300251322217
     },
     {
       "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": 19004.70022508996,
-      "required_pages_per_second_upper_estimate": 0.004396402925774214
+      "required_token_positions_per_second": 18916.325400290672,
+      "required_pages_per_second_upper_estimate": 0.0043759589654006935
     },
     {
       "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": 28507.05033763494,
-      "required_pages_per_second_upper_estimate": 0.0017359298006278335
+      "required_token_positions_per_second": 28374.48810043601,
+      "required_pages_per_second_upper_estimate": 0.001727857455882726
     },
     {
       "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": 28507.05033763494,
-      "required_pages_per_second_upper_estimate": 0.002194551647557101
+      "required_token_positions_per_second": 28374.48810043601,
+      "required_pages_per_second_upper_estimate": 0.002184346639581769
     },
     {
       "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": 28507.05033763494,
-      "required_pages_per_second_upper_estimate": 0.35396215823329574
+      "required_token_positions_per_second": 28374.48810043601,
+      "required_pages_per_second_upper_estimate": 0.3523161789045533
     }
   ],
   "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": 0.0794873151563138,
-    "producer_p95_seconds": 0.3239012379999622,
-    "gpu_update_seconds": 0.9698653376108568,
-    "producer_batches_per_second": 12.580623688615912,
-    "gpu_batches_per_second": 1.0310709757535785,
-    "producer_headroom": 12.201510841114615,
+    "producer_batch_seconds": 0.13727578607029045,
+    "producer_p95_seconds": 0.3475367910032219,
+    "gpu_update_seconds": 0.9743964332373332,
+    "producer_batches_per_second": 7.284605891733607,
+    "gpu_batches_per_second": 1.0262763346511867,
+    "producer_headroom": 7.098093998444889,
     "sustainable": true,
-    "buffer_seconds": 60.131650931873125,
+    "buffer_seconds": 60.41257886071465,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 406480.6035923726,
-    "upload_bytes_per_second": 5537825.4582211785,
+    "generation_bytes_per_second": 481137.92799454666,
+    "upload_bytes_per_second": 5487182.841082292,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.22127514675506,
+    "estimated_full_shards_per_hour": 103.24099902989673,
     "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": 1.1522124919974885,
-      "bytes": 8515706,
-      "pages": 41,
-      "shards": 4
-    },
     {
       "seconds": 1.1056661020011234,
       "bytes": 6106174,
@@ -562,6 +556,12 @@
       "bytes": 6229619,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.705037297000672,
+      "bytes": 7666167,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -246,3 +246,4 @@
 {"step": 179600, "loss": 0.6041311882436275, "lr": 0.0003, "grad_norm": 0.1311485899855217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26285640226, "tokens_per_second": 148634.68012424355, "data_wait_s": 1.1896998330485076e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 179700, "loss": 0.6336561497300863, "lr": 0.0003, "grad_norm": 0.17994062408677464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26300344740, "tokens_per_second": 150434.4264269306, "data_wait_s": 8.998998964671046e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 179800, "loss": 0.5804962553083897, "lr": 0.0003, "grad_norm": 0.15476970584776104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26315108925, "tokens_per_second": 149339.41115634065, "data_wait_s": 1.3672000932274386e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 179900, "loss": 0.6551252659410238, "lr": 0.0003, "grad_norm": 0.1586618530744718, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26329795713, "tokens_per_second": 149333.07682973368, "data_wait_s": 1.1494001228129491e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:24e1d32faad1f854f7e9fca6618fc3c520743a34038716824dd89e33949ec843
-size 528561
+oid sha256:5981219f613ed52b36d14cbf59753d4eba7263fb497acc27f77194932e186865
+size 530679