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": 76431.70005325026,
-      "required_pages_per_second_upper_estimate": 0.005287722355661484
+      "required_token_positions_per_second": 77025.62074989194,
+      "required_pages_per_second_upper_estimate": 0.005328811167541035
     },
     {
       "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": 9553.962506656282,
-      "required_pages_per_second_upper_estimate": 0.004000044592112874
+      "required_token_positions_per_second": 9628.202593736492,
+      "required_pages_per_second_upper_estimate": 0.0040311273662640475
     },
     {
       "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": 19107.925013312564,
-      "required_pages_per_second_upper_estimate": 0.003979676870403935
+      "required_token_positions_per_second": 19256.405187472985,
+      "required_pages_per_second_upper_estimate": 0.004010601374996039
     },
     {
       "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": 28661.887519968845,
-      "required_pages_per_second_upper_estimate": 0.0017182363926539167
+      "required_token_positions_per_second": 28884.607781209477,
+      "required_pages_per_second_upper_estimate": 0.0017315881322411447
     },
     {
       "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": 28661.887519968845,
-      "required_pages_per_second_upper_estimate": 0.0022283356843578995
+      "required_token_positions_per_second": 28884.607781209477,
+      "required_pages_per_second_upper_estimate": 0.002245651205026461
     },
     {
       "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": 28661.887519968845,
-      "required_pages_per_second_upper_estimate": 0.3676439183690415
+      "required_token_positions_per_second": 28884.607781209477,
+      "required_pages_per_second_upper_estimate": 0.3705007347418514
     }
   ],
   "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": 0.10626844970320803,
-    "producer_p95_seconds": 0.2982578849987476,
-    "gpu_update_seconds": 0.9646259333317645,
-    "producer_batches_per_second": 9.410130690650435,
-    "gpu_batches_per_second": 1.0366712789340584,
-    "producer_headroom": 9.077256100242558,
+    "producer_batch_seconds": 0.2129644436563467,
+    "producer_p95_seconds": 0.28998892399977194,
+    "gpu_update_seconds": 0.957188001631307,
+    "producer_batches_per_second": 4.695619526110495,
+    "gpu_batches_per_second": 1.0447268439384214,
+    "producer_headroom": 4.49459067061865,
     "sustainable": true,
-    "buffer_seconds": 60.77143379990116,
+    "buffer_seconds": 60.30284410277234,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 347661.94227066403,
-    "upload_bytes_per_second": 5080571.87659806,
+    "generation_bytes_per_second": 341384.48172185593,
+    "upload_bytes_per_second": 5169525.167320821,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.60015965547505,
+    "estimated_full_shards_per_hour": 73.2531627535034,
     "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": 1.439283288000297,
-      "bytes": 6018491,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.925738228001137,
       "bytes": 863516,
@@ -562,6 +556,12 @@
       "bytes": 4885086,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.4905629579989181,
+      "bytes": 13725537,
+      "pages": 44,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -273,3 +273,4 @@
 {"step": 952300, "loss": 0.5251018404960632, "lr": 3e-05, "grad_norm": 0.34399387024216427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139390817641, "tokens_per_second": 152716.52592594008, "data_wait_s": 1.2970998795935884e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 952400, "loss": 0.5081045627593994, "lr": 3e-05, "grad_norm": 0.3509164334995831, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139405519742, "tokens_per_second": 152630.40931622678, "data_wait_s": 1.0414998541818932e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 952500, "loss": 0.5086105465888977, "lr": 3e-05, "grad_norm": 0.3440548034632097, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139420057862, "tokens_per_second": 150938.92247970775, "data_wait_s": 1.4551002095686272e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 952600, "loss": 0.5340831279754639, "lr": 3e-05, "grad_norm": 0.3493283224978132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139434654639, "tokens_per_second": 151007.9218448461, "data_wait_s": 1.3346001651370898e-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 sha256:95907461e3aa0cdafe09943e0252c987af1473c4d46e0c47488b42fd9dd0560e
-size 586530
+oid sha256:9725cd749cb10db232aa57f8ac03b636785e37187760a1ef38e4fae3897a0dfa
+size 588648