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": 77274.69984608729,
-      "required_pages_per_second_upper_estimate": 0.005346043037358957
+      "required_token_positions_per_second": 77194.4574599123,
+      "required_pages_per_second_upper_estimate": 0.005340491682895384
     },
     {
       "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": 9659.33748076091,
-      "required_pages_per_second_upper_estimate": 0.004044162893290797
+      "required_token_positions_per_second": 9649.307182489038,
+      "required_pages_per_second_upper_estimate": 0.00403996341686081
     },
     {
       "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": 19318.67496152182,
-      "required_pages_per_second_upper_estimate": 0.00402357052676604
+      "required_token_positions_per_second": 19298.614364978075,
+      "required_pages_per_second_upper_estimate": 0.004019392433539484
     },
     {
       "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": 28978.012442282732,
-      "required_pages_per_second_upper_estimate": 0.0017371875990518269
+      "required_token_positions_per_second": 28947.92154746711,
+      "required_pages_per_second_upper_estimate": 0.0017353836958537664
     },
     {
       "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": 28978.012442282732,
-      "required_pages_per_second_upper_estimate": 0.002252913006581222
+      "required_token_positions_per_second": 28947.92154746711,
+      "required_pages_per_second_upper_estimate": 0.002250573571865166
     },
     {
       "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": 28978.012442282732,
-      "required_pages_per_second_upper_estimate": 0.3716988294439878
+      "required_token_positions_per_second": 28947.92154746711,
+      "required_pages_per_second_upper_estimate": 0.3713128557543786
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.11258031489859377,
-    "producer_p95_seconds": 0.19026688499980082,
-    "gpu_update_seconds": 0.9541027030431505,
-    "producer_batches_per_second": 8.882547547506379,
-    "gpu_batches_per_second": 1.048105195395064,
-    "producer_headroom": 8.474862624985144,
+    "producer_batch_seconds": 0.2308806265780987,
+    "producer_p95_seconds": 0.20477265199951944,
+    "gpu_update_seconds": 0.9550944773241984,
+    "producer_batches_per_second": 4.33124257682892,
+    "gpu_batches_per_second": 1.0470168383777168,
+    "producer_headroom": 4.136745865080732,
     "sustainable": true,
-    "buffer_seconds": 60.10847029171848,
+    "buffer_seconds": 60.1709520714245,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 531842.340831932,
-    "upload_bytes_per_second": 5779482.881375651,
+    "generation_bytes_per_second": 528130.9429880334,
+    "upload_bytes_per_second": 5669940.383023089,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.12098568647833,
+    "estimated_full_shards_per_hour": 113.32460610609772,
     "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": 1.431994874999873,
-      "bytes": 7632897,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.237401497999599,
       "bytes": 5052734,
@@ -562,6 +556,12 @@
       "bytes": 6027470,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.9445103569996718,
+      "bytes": 2479005,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -114,3 +114,4 @@
 {"step": 536400, "loss": 0.5694395899772644, "lr": 0.0003, "grad_norm": 0.17182849504903272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78511539349, "tokens_per_second": 153951.41072311337, "data_wait_s": 1.1206000635866076e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 536500, "loss": 0.5985449552536011, "lr": 0.0003, "grad_norm": 0.1573570286032401, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78526239509, "tokens_per_second": 153347.29160659207, "data_wait_s": 1.7140999261755496e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 536600, "loss": 0.5950390696525574, "lr": 0.0003, "grad_norm": 0.1656959413923456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78540915553, "tokens_per_second": 153613.49640257735, "data_wait_s": 1.2935000995639712e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 536700, "loss": 0.659512996673584, "lr": 0.0003, "grad_norm": 0.15999194652839047, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78555528991, "tokens_per_second": 152619.0411828723, "data_wait_s": 1.0633999409037642e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4244409156bb4300de2e643a3dba480a35101ca5aada364bf38fad1b1f97d2fc
-size 247419
+oid sha256:0bc5026ea212174b5e58cdd541265b6b2bbfaacbb08771d2fc867934732e3444
+size 249537