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": 75335.00870894277,
-      "required_pages_per_second_upper_estimate": 0.0065298476200571975
+      "required_token_positions_per_second": 74340.72972774111,
+      "required_pages_per_second_upper_estimate": 0.006443666037943677
     },
     {
       "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": 9416.876088617846,
-      "required_pages_per_second_upper_estimate": 0.0037565326665940026
+      "required_token_positions_per_second": 9292.591215967639,
+      "required_pages_per_second_upper_estimate": 0.003706953572669395
     },
     {
       "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": 18833.752177235692,
-      "required_pages_per_second_upper_estimate": 0.004356857103486015
+      "required_token_positions_per_second": 18585.182431935278,
+      "required_pages_per_second_upper_estimate": 0.004299354867588868
     },
     {
       "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": 28250.628265853535,
-      "required_pages_per_second_upper_estimate": 0.001720315041798983
+      "required_token_positions_per_second": 27877.773647902915,
+      "required_pages_per_second_upper_estimate": 0.0016976101517827862
     },
     {
       "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": 28250.628265853535,
-      "required_pages_per_second_upper_estimate": 0.0021748115666496435
+      "required_token_positions_per_second": 27877.773647902915,
+      "required_pages_per_second_upper_estimate": 0.002146108186032163
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28250.628265853535,
-      "required_pages_per_second_upper_estimate": 0.35077825429123927
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27877.773647902915,
+      "required_pages_per_second_upper_estimate": 0.3461486478004261
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09414915059373996,
-    "producer_p95_seconds": 0.16845601699969848,
-    "gpu_update_seconds": 0.9786685003893548,
-    "producer_batches_per_second": 10.621444736289428,
-    "gpu_batches_per_second": 1.0217964505878738,
-    "producer_headroom": 10.394873392032782,
+    "producer_batch_seconds": 0.20213575359375113,
+    "producer_p95_seconds": 0.17468009399999573,
+    "gpu_update_seconds": 0.9917578193006026,
+    "producer_batches_per_second": 4.947170316092532,
+    "gpu_batches_per_second": 1.008310678815933,
+    "producer_headroom": 4.906394844396602,
     "sustainable": true,
-    "buffer_seconds": 60.67744702414,
+    "buffer_seconds": 60.49722697733676,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 162,
+    "upload_pages_per_commit": 156,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 606365.3590635646,
-    "upload_bytes_per_second": 5666132.37370981,
+    "generation_bytes_per_second": 587684.2498395673,
+    "upload_bytes_per_second": 5665389.843365814,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.05594529595471,
+    "estimated_full_shards_per_hour": 63.05167971320279,
     "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": [
@@ -346,6 +346,12 @@
       "bytes": 7523243,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.8447506450002038,
+      "bytes": 4764163,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -30,3 +30,4 @@
 {"step": 84900, "loss": 0.6152340620756149, "lr": 0.0003, "grad_norm": 0.12822516601325828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12424713818, "tokens_per_second": 147683.3784253226, "data_wait_s": 6.062000011297641e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 85000, "loss": 0.6123912185430527, "lr": 0.0003, "grad_norm": 0.12991575957755994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12439396796, "tokens_per_second": 148490.2662518952, "data_wait_s": 1.2061000234098174e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 85100, "loss": 0.6527527756989002, "lr": 0.0003, "grad_norm": 0.1299015585553546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12454100277, "tokens_per_second": 134800.67630284632, "data_wait_s": 9.854999916569795e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 85200, "loss": 0.602685734629631, "lr": 0.0003, "grad_norm": 0.13874306371124315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12468770344, "tokens_per_second": 148680.2669662592, "data_wait_s": 1.0623999969539e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:82a177f24055d8438d5870cf82b0289306de6f5aefd2c84d2d8a35f2e10dfa44
-size 68724
+oid sha256:01f41aff2ce7ba40d890aa799299aba352e99f4c2dbe9f4341ee2d637e778200
+size 70842