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": 75195.57683476034,
-      "required_pages_per_second_upper_estimate": 0.006517762018589919
+      "required_token_positions_per_second": 75781.88982551989,
+      "required_pages_per_second_upper_estimate": 0.006568582142632267
     },
     {
       "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": 9399.447104345043,
-      "required_pages_per_second_upper_estimate": 0.0037495799841810447
+      "required_token_positions_per_second": 9472.736228189986,
+      "required_pages_per_second_upper_estimate": 0.0037788161114528426
     },
     {
       "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": 18798.894208690086,
-      "required_pages_per_second_upper_estimate": 0.004348793325942283
+      "required_token_positions_per_second": 18945.472456379972,
+      "required_pages_per_second_upper_estimate": 0.004382701623856286
     },
     {
       "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": 28198.34131303513,
-      "required_pages_per_second_upper_estimate": 0.0017171310407007822
+      "required_token_positions_per_second": 28418.208684569956,
+      "required_pages_per_second_upper_estimate": 0.0017305198100723061
     },
     {
       "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": 28198.34131303513,
-      "required_pages_per_second_upper_estimate": 0.002170786372282133
+      "required_token_positions_per_second": 28418.208684569956,
+      "required_pages_per_second_upper_estimate": 0.002187712371174723
     },
     {
       "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": 28198.34131303513,
-      "required_pages_per_second_upper_estimate": 0.35012902533040874
+      "required_token_positions_per_second": 28418.208684569956,
+      "required_pages_per_second_upper_estimate": 0.35285904223611453
     }
   ],
   "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": 0.38839714217967014,
-    "producer_p95_seconds": 0.31320281800071825,
-    "gpu_update_seconds": 0.9804832026492024,
-    "producer_batches_per_second": 2.5746842378603447,
-    "gpu_batches_per_second": 1.019905284759662,
-    "producer_headroom": 2.5244346473477317,
+    "producer_batch_seconds": 0.06408935626562595,
+    "producer_p95_seconds": 0.21530763400005526,
+    "gpu_update_seconds": 0.972897352781136,
+    "producer_batches_per_second": 15.60321492160697,
+    "gpu_batches_per_second": 1.027857663649087,
+    "producer_headroom": 15.180326492106541,
     "sustainable": true,
-    "buffer_seconds": 60.78995856425055,
+    "buffer_seconds": 60.31963587243043,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 395445.4120105795,
-    "upload_bytes_per_second": 5251945.3196084,
+    "generation_bytes_per_second": 395541.76703755104,
+    "upload_bytes_per_second": 5188295.0534078,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.85337991941489,
+    "estimated_full_shards_per_hour": 84.87405546517275,
     "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": 1.3391879739992874,
-      "bytes": 13717263,
-      "pages": 44,
-      "shards": 4
-    },
     {
       "seconds": 0.882563871999082,
       "bytes": 1071171,
@@ -562,6 +556,12 @@
       "bytes": 6338726,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.157190105999689,
+      "bytes": 7618038,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -156,3 +156,4 @@
 {"step": 170600, "loss": 0.5806423630565405, "lr": 0.0003, "grad_norm": 0.1279877202192422, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24968105872, "tokens_per_second": 148764.0496575284, "data_wait_s": 1.5081001038197428e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 170700, "loss": 0.5654618572443724, "lr": 0.0003, "grad_norm": 0.13484862298589473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24982779626, "tokens_per_second": 148712.44759760736, "data_wait_s": 1.456300014979206e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 170800, "loss": 0.5619199220091105, "lr": 0.0003, "grad_norm": 0.1262704703167331, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24997567974, "tokens_per_second": 151503.7682091754, "data_wait_s": 1.06999996205559e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 170900, "loss": 0.577710309997201, "lr": 0.0003, "grad_norm": 0.16445502639002604, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25012322021, "tokens_per_second": 151245.76910402786, "data_wait_s": 1.3898999895900488e-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 sha256:8c2eea9b2b7094c45485b3bdf05f20dcfc1e2a193c2bffa85f4e8b4fc0a1e149
-size 337158
+oid sha256:f824adcf8424dffa39fb8f322b9e1ccde62392d9e972cd165afffb7eed82c053
+size 339276