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": 77215.51033368144,
-      "required_pages_per_second_upper_estimate": 0.00534194816955213
+      "required_token_positions_per_second": 77211.50181224127,
+      "required_pages_per_second_upper_estimate": 0.005341670850737848
     },
     {
       "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": 9651.93879171018,
-      "required_pages_per_second_upper_estimate": 0.004041065216687453
+      "required_token_positions_per_second": 9651.43772653016,
+      "required_pages_per_second_upper_estimate": 0.004040855431160009
     },
     {
       "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": 19303.87758342036,
-      "required_pages_per_second_upper_estimate": 0.004020488623140608
+      "required_token_positions_per_second": 19302.87545306032,
+      "required_pages_per_second_upper_estimate": 0.0040202799058145664
     },
     {
       "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": 28955.816375130536,
-      "required_pages_per_second_upper_estimate": 0.001735856978717485
+      "required_token_positions_per_second": 28954.313179590474,
+      "required_pages_per_second_upper_estimate": 0.001735766864440105
     },
     {
       "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": 28955.816375130536,
-      "required_pages_per_second_upper_estimate": 0.0022511873599903205
+      "required_token_positions_per_second": 28954.313179590474,
+      "required_pages_per_second_upper_estimate": 0.0022510704931489388
     },
     {
       "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": 28955.816375130536,
-      "required_pages_per_second_upper_estimate": 0.37141412212683955
+      "required_token_positions_per_second": 28954.313179590474,
+      "required_pages_per_second_upper_estimate": 0.37139484074845724
     }
   ],
   "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": 0.10153320464866056,
-    "producer_p95_seconds": 0.35486642199975904,
-    "gpu_update_seconds": 0.9548340700124832,
-    "producer_batches_per_second": 9.848994754576498,
-    "gpu_batches_per_second": 1.0473023862532747,
-    "producer_headroom": 9.404155747043877,
+    "producer_batch_seconds": 0.11919808033582058,
+    "producer_p95_seconds": 0.37159141700249165,
+    "gpu_update_seconds": 0.9548836412907463,
+    "producer_batches_per_second": 8.389396852555578,
+    "gpu_batches_per_second": 1.0472480172016232,
+    "producer_headroom": 8.010897814801396,
     "sustainable": true,
-    "buffer_seconds": 60.154546410786445,
+    "buffer_seconds": 60.157669401317015,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 423278.3939063002,
-    "upload_bytes_per_second": 5041480.47465339,
+    "generation_bytes_per_second": 425178.4183204969,
+    "upload_bytes_per_second": 5072060.937761938,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.82568991557841,
+    "estimated_full_shards_per_hour": 91.23339092456037,
     "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": 1.757866293999541,
-      "bytes": 6331179,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.503990763994807,
       "bytes": 11868685,
@@ -562,6 +556,12 @@
       "bytes": 13518430,
       "pages": 43,
       "shards": 4
+    },
+    {
+      "seconds": 1.2164938489950146,
+      "bytes": 6185487,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -411,3 +411,4 @@
 {"step": 966100, "loss": 0.5369018912315369, "lr": 3e-05, "grad_norm": 0.37190364193231235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141410572187, "tokens_per_second": 153852.86583037066, "data_wait_s": 7.890994311310351e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 966200, "loss": 0.546553909778595, "lr": 3e-05, "grad_norm": 0.3533076130064056, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141425317778, "tokens_per_second": 153513.28357493415, "data_wait_s": 1.846899976953864e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 966300, "loss": 0.5306552052497864, "lr": 3e-05, "grad_norm": 0.355788559411236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141440034338, "tokens_per_second": 153460.48339376223, "data_wait_s": 1.8106999050360173e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 966400, "loss": 0.5564902424812317, "lr": 3e-05, "grad_norm": 0.36915678641367994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141454611219, "tokens_per_second": 151082.07039932828, "data_wait_s": 2.622199826873839e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:61d9d7d330670bc231bd13b91b5fda3c493ae09df80496f702b3d2f3c54281e5
-size 881163
+oid sha256:02127f2d43f04c1b99cc5a58906ace04307bcbc5358cfb89257e7cad3b2aa88f
+size 883281