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": 74774.38045179083,
-      "required_pages_per_second_upper_estimate": 0.006481253783626594
+      "required_token_positions_per_second": 75498.36329245908,
+      "required_pages_per_second_upper_estimate": 0.0065440067813379845
     },
     {
       "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": 9346.797556473854,
-      "required_pages_per_second_upper_estimate": 0.0037285772923543376
+      "required_token_positions_per_second": 9437.295411557385,
+      "required_pages_per_second_upper_estimate": 0.0037646782398106688
     },
     {
       "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": 18693.595112947707,
-      "required_pages_per_second_upper_estimate": 0.004324434233343074
+      "required_token_positions_per_second": 18874.59082311477,
+      "required_pages_per_second_upper_estimate": 0.004366304405474521
     },
     {
       "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": 28040.39266942156,
-      "required_pages_per_second_upper_estimate": 0.0017075127969972041
+      "required_token_positions_per_second": 28311.886234672158,
+      "required_pages_per_second_upper_estimate": 0.0017240453307043135
     },
     {
       "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": 28040.39266942156,
-      "required_pages_per_second_upper_estimate": 0.002158627048466931
+      "required_token_positions_per_second": 28311.886234672158,
+      "required_pages_per_second_upper_estimate": 0.002179527374662212
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28040.39266942156,
-      "required_pages_per_second_upper_estimate": 0.34816783179683325
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28311.886234672158,
+      "required_pages_per_second_upper_estimate": 0.3515388732467333
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.29991811579708383,
-    "producer_p95_seconds": 0.3645642640040023,
-    "gpu_update_seconds": 0.9860061635353107,
-    "producer_batches_per_second": 3.334243406212154,
-    "gpu_batches_per_second": 1.0141924431937774,
-    "producer_headroom": 3.2875845492521525,
+    "producer_batch_seconds": 0.09600396925793575,
+    "producer_p95_seconds": 0.36925291599618504,
+    "gpu_update_seconds": 0.9765509712362743,
+    "producer_batches_per_second": 10.416235992423191,
+    "gpu_batches_per_second": 1.0240120889276676,
+    "producer_headroom": 10.171985375027106,
     "sustainable": true,
-    "buffer_seconds": 60.14637597565395,
+    "buffer_seconds": 60.54616021664901,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 332471.5637150459,
-    "upload_bytes_per_second": 5209264.115662761,
+    "generation_bytes_per_second": 364586.4794256969,
+    "upload_bytes_per_second": 5248603.161211893,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.34065803135427,
+    "estimated_full_shards_per_hour": 78.23177134588413,
     "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": 0.9964947309999843,
-      "bytes": 5697842,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.1368190850007522,
       "bytes": 10318857,
@@ -562,6 +556,12 @@
       "bytes": 1251369,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.8413397670010454,
+      "bytes": 13265988,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -382,3 +382,4 @@
 {"step": 275300, "loss": 0.571684068068862, "lr": 0.0003, "grad_norm": 0.13297020996572717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40295403551, "tokens_per_second": 150699.85841685484, "data_wait_s": 1.8714999896474183e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 275400, "loss": 0.5842844024300575, "lr": 0.0003, "grad_norm": 0.15522975680030643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40310118451, "tokens_per_second": 150884.40286893057, "data_wait_s": 1.2607000826392323e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 275500, "loss": 0.6546776555478573, "lr": 0.0003, "grad_norm": 0.15023174520998175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40324827474, "tokens_per_second": 149574.15061896693, "data_wait_s": 1.3569006114266813e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 275600, "loss": 0.6014276780188084, "lr": 0.0003, "grad_norm": 0.14329618538429756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40339605846, "tokens_per_second": 150646.99400476814, "data_wait_s": 1.5729005099274218e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5408f67de1acc8b6be333bad867e2857b555418d8fc6482d43dbac4a9b762724
-size 819741
+oid sha256:de914c731b928565c99764c7888ec5703475a80b504d25aa0d95334a2ad1502a
+size 821859