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": 75614.95079605276,
-      "required_pages_per_second_upper_estimate": 0.006554112290661158
+      "required_token_positions_per_second": 73518.29645419325,
+      "required_pages_per_second_upper_estimate": 0.006372379606230623
     },
     {
       "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": 9451.868849506594,
-      "required_pages_per_second_upper_estimate": 0.0037704918021009234
+      "required_token_positions_per_second": 9189.787056774156,
+      "required_pages_per_second_upper_estimate": 0.0036659434565079604
     },
     {
       "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": 18903.73769901319,
-      "required_pages_per_second_upper_estimate": 0.004373047022246124
+      "required_token_positions_per_second": 18379.57411354831,
+      "required_pages_per_second_upper_estimate": 0.004251791001712857
     },
     {
       "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": 28355.606548519783,
-      "required_pages_per_second_upper_estimate": 0.0017267076684348748
+      "required_token_positions_per_second": 27569.361170322467,
+      "required_pages_per_second_upper_estimate": 0.0016788294500133523
     },
     {
       "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": 28355.606548519783,
-      "required_pages_per_second_upper_estimate": 0.002182893085447777
+      "required_token_positions_per_second": 27569.361170322467,
+      "required_pages_per_second_upper_estimate": 0.0021223657397676406
     },
     {
       "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": 28355.606548519783,
-      "required_pages_per_second_upper_estimate": 0.352081733222243
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27569.361170322467,
+      "required_pages_per_second_upper_estimate": 0.3423191970190405
     }
   ],
   "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": 62,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.09588674500011507,
-    "producer_p95_seconds": 0.24152096300167614,
-    "gpu_update_seconds": 0.9750452684794809,
-    "producer_batches_per_second": 10.42897013553646,
-    "gpu_batches_per_second": 1.025593408149587,
-    "producer_headroom": 10.168717985768636,
+    "producer_batch_seconds": 0.5595638596016954,
+    "producer_p95_seconds": 0.30422171100144624,
+    "gpu_update_seconds": 1.0028523994151226,
+    "producer_batches_per_second": 1.7871061235295156,
+    "gpu_batches_per_second": 0.997155713625668,
+    "producer_headroom": 1.7922036639910333,
     "sustainable": true,
-    "buffer_seconds": 60.45280664572782,
+    "buffer_seconds": 60.17114396490736,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 438870.18220032955,
-    "upload_bytes_per_second": 5755016.218284796,
+    "generation_bytes_per_second": 379580.7000230582,
+    "upload_bytes_per_second": 5843493.273345275,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.17132472522178,
+    "estimated_full_shards_per_hour": 81.44918203848657,
     "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": 1.2737176890004775,
-      "bytes": 7770049,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.1648699769993982,
       "bytes": 6106287,
@@ -562,6 +556,12 @@
       "bytes": 1089066,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.0898243610026839,
+      "bytes": 13191576,
+      "pages": 31,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -194,3 +194,4 @@
 {"step": 374400, "loss": 0.5192107986658812, "lr": 0.0003, "grad_norm": 0.15734461036322073, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54801338518, "tokens_per_second": 147650.84406482143, "data_wait_s": 1.2248001439729705e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 374500, "loss": 0.663544163107872, "lr": 0.0003, "grad_norm": 0.17155917032009105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54816014619, "tokens_per_second": 149832.33806820677, "data_wait_s": 1.251099820365198e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 374600, "loss": 0.6049759332090616, "lr": 0.0003, "grad_norm": 0.17264880688967682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54830715951, "tokens_per_second": 149078.2490447577, "data_wait_s": 1.0477000614628196e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 374700, "loss": 0.5894653387367725, "lr": 0.0003, "grad_norm": 0.1596533167155301, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54845284765, "tokens_per_second": 148263.97386253468, "data_wait_s": 1.0391002433607355e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:09b7cf257e9a3c719540921e6d6f952c519b2b8f89413d0b67e42d3da71986e1
-size 417642
+oid sha256:4f087e97d27caeb27ccd1c45d158e36bdfdd9eaceec085ba904d051ec69715f1
+size 419760