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": 76624.6568465838,
-      "required_pages_per_second_upper_estimate": 0.005301071554345743
+      "required_token_positions_per_second": 76421.92807386078,
+      "required_pages_per_second_upper_estimate": 0.005287046307453204
     },
     {
       "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": 9578.082105822976,
-      "required_pages_per_second_upper_estimate": 0.004010142964609463
+      "required_token_positions_per_second": 9552.741009232597,
+      "required_pages_per_second_upper_estimate": 0.003999533176649343
     },
     {
       "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": 19156.16421164595,
-      "required_pages_per_second_upper_estimate": 0.003989723823263571
+      "required_token_positions_per_second": 19105.482018465194,
+      "required_pages_per_second_upper_estimate": 0.003979168059003334
     },
     {
       "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": 28734.24631746893,
-      "required_pages_per_second_upper_estimate": 0.0017225741920785605
+      "required_token_positions_per_second": 28658.223027697793,
+      "required_pages_per_second_upper_estimate": 0.001718016711937623
     },
     {
       "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": 28734.24631746893,
-      "required_pages_per_second_upper_estimate": 0.0022339612625908167
+      "required_token_positions_per_second": 28658.223027697793,
+      "required_pages_per_second_upper_estimate": 0.002228050786202223
     },
     {
       "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": 28734.24631746893,
-      "required_pages_per_second_upper_estimate": 0.368572059330549
+      "required_token_positions_per_second": 28658.223027697793,
+      "required_pages_per_second_upper_estimate": 0.36759691419681373
     }
   ],
   "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": 0.48308491897651606,
-    "producer_p95_seconds": 0.27002714300033404,
-    "gpu_update_seconds": 0.9621968049738424,
-    "producer_batches_per_second": 2.0700294310959695,
-    "gpu_batches_per_second": 1.0392884229408612,
-    "producer_headroom": 1.9917757048023623,
+    "producer_batch_seconds": 0.09368686961715866,
+    "producer_p95_seconds": 0.28088526699866634,
+    "gpu_update_seconds": 0.9647492788816171,
+    "producer_batches_per_second": 10.67385434145033,
+    "gpu_batches_per_second": 1.0365387379809676,
+    "producer_headroom": 10.297593278801624,
     "sustainable": true,
-    "buffer_seconds": 60.61839871335207,
+    "buffer_seconds": 60.779204569541875,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 290969.72877663944,
-    "upload_bytes_per_second": 5293873.765262302,
+    "generation_bytes_per_second": 291563.700715511,
+    "upload_bytes_per_second": 5283842.287151126,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.435330366844056,
+    "estimated_full_shards_per_hour": 62.562782917966814,
     "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": 1.3083442860006471,
-      "bytes": 8430845,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0249396090002847,
       "bytes": 5941897,
@@ -562,6 +556,12 @@
       "bytes": 6177508,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.05712037199919,
+      "bytes": 6292479,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -219,3 +219,4 @@
 {"step": 626300, "loss": 0.5707839727401733, "lr": 0.0001, "grad_norm": 0.16334502314435753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91674039090, "tokens_per_second": 151943.28471567607, "data_wait_s": 1.7668997315922752e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 626400, "loss": 0.5396880507469177, "lr": 0.0001, "grad_norm": 0.14936795053782706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91688532080, "tokens_per_second": 150298.5900370969, "data_wait_s": 1.7931997717823833e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 626500, "loss": 0.5421844124794006, "lr": 0.0001, "grad_norm": 0.16092432121913808, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91703019884, "tokens_per_second": 138728.96300380188, "data_wait_s": 1.2681000953307375e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 626600, "loss": 0.5956579446792603, "lr": 0.0001, "grad_norm": 0.17662651419162206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91717627485, "tokens_per_second": 150343.20490274075, "data_wait_s": 1.5474997780984268e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:67fc25c3b8df3021c728bf15d18c5018b8d6620792afbd6f459d71b653bcebde
-size 472158
+oid sha256:5f27c6f2860e23a1f87cd658dd4a74cdddf44109bb5eadabecbb24d535e90d9b
+size 474276