Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77115.05252227574,
-      "required_pages_per_second_upper_estimate": 0.005334998265064847
+      "required_token_positions_per_second": 76745.22415520168,
+      "required_pages_per_second_upper_estimate": 0.0053094126805106265
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9639.381565284468,
-      "required_pages_per_second_upper_estimate": 0.004035807768207713
+      "required_token_positions_per_second": 9593.15301940021,
+      "required_pages_per_second_upper_estimate": 0.004016452841407787
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19278.763130568936,
-      "required_pages_per_second_upper_estimate": 0.004015257944924316
+      "required_token_positions_per_second": 19186.30603880042,
+      "required_pages_per_second_upper_estimate": 0.00399600157096641
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28918.144695853403,
-      "required_pages_per_second_upper_estimate": 0.0017335986190661445
+      "required_token_positions_per_second": 28779.459058200628,
+      "required_pages_per_second_upper_estimate": 0.001725284626849568
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28918.144695853403,
-      "required_pages_per_second_upper_estimate": 0.0022482585526267286
+      "required_token_positions_per_second": 28779.459058200628,
+      "required_pages_per_second_upper_estimate": 0.002237476354312877
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28918.144695853403,
-      "required_pages_per_second_upper_estimate": 0.3709309102737703
+      "required_token_positions_per_second": 28779.459058200628,
+      "required_pages_per_second_upper_estimate": 0.36915199982299646
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13186199204683646,
-    "producer_p95_seconds": 0.2106983779995062,
-    "gpu_update_seconds": 0.9560779327577149,
-    "producer_batches_per_second": 7.583686432135857,
-    "gpu_batches_per_second": 1.0459398399831237,
-    "producer_headroom": 7.250595246719181,
+    "producer_batch_seconds": 0.0896960804531517,
+    "producer_p95_seconds": 0.203442788000757,
+    "gpu_update_seconds": 0.9606851867537718,
+    "producer_batches_per_second": 11.148759175963107,
+    "gpu_batches_per_second": 1.040923721723113,
+    "producer_headroom": 10.710447791032944,
     "sustainable": true,
-    "buffer_seconds": 60.23290976373604,
+    "buffer_seconds": 60.52316676548762,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344001.31929271173,
-    "upload_bytes_per_second": 4204762.160072704,
+    "generation_bytes_per_second": 344217.13167753734,
+    "upload_bytes_per_second": 4198465.83003919,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.81467517934813,
+    "estimated_full_shards_per_hour": 73.8609834932765,
     "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": 1.5345939340004406,
-      "bytes": 10306218,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.8383978550000393,
       "bytes": 2628306,
@@ -562,6 +556,12 @@
       "bytes": 5205900,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.034200502999738,
+      "bytes": 7563521,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -142,3 +142,4 @@
 {"step": 699200, "loss": 0.5393723845481873, "lr": 8e-05, "grad_norm": 0.22657107023763956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102343932128, "tokens_per_second": 153355.68077422585, "data_wait_s": 1.569199957884848e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 699300, "loss": 0.533176600933075, "lr": 8e-05, "grad_norm": 0.20982104424504638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102358416299, "tokens_per_second": 151209.50613614768, "data_wait_s": 8.457998774247244e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 699400, "loss": 0.5447697043418884, "lr": 8e-05, "grad_norm": 0.20386226727672935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102372998603, "tokens_per_second": 151799.83233677337, "data_wait_s": 1.4466000720858574e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 699500, "loss": 0.5255951285362244, "lr": 8e-05, "grad_norm": 0.20401954144534976, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102387638859, "tokens_per_second": 153350.35853250782, "data_wait_s": 1.0024999937741086e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:33db246e47a4803f6d049e2d048077a8baa4422d6021e596a5d94a5f11a766a0
-size 306723
+oid sha256:d64013642382dfc0812db92adac874bd8ba08ba0f7d73966e072219d50c17669
+size 308841