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": 76721.83610707126,
-      "required_pages_per_second_upper_estimate": 0.005307794641073216
+      "required_token_positions_per_second": 76782.88124573114,
+      "required_pages_per_second_upper_estimate": 0.005312017885409943
     },
     {
       "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": 9590.229513383907,
-      "required_pages_per_second_upper_estimate": 0.00401522883048851
+      "required_token_positions_per_second": 9597.860155716393,
+      "required_pages_per_second_upper_estimate": 0.00401842362108719
     },
     {
       "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": 19180.459026767814,
-      "required_pages_per_second_upper_estimate": 0.003994783792556095
+      "required_token_positions_per_second": 19195.720311432786,
+      "required_pages_per_second_upper_estimate": 0.0039979623156846675
     },
     {
       "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": 28770.68854015172,
-      "required_pages_per_second_upper_estimate": 0.0017247588476843416
+      "required_token_positions_per_second": 28793.58046714918,
+      "required_pages_per_second_upper_estimate": 0.0017261311837538907
     },
     {
       "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": 28770.68854015172,
-      "required_pages_per_second_upper_estimate": 0.0022367944851120064
+      "required_token_positions_per_second": 28793.58046714918,
+      "required_pages_per_second_upper_estimate": 0.0022385742317451132
     },
     {
       "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": 28770.68854015172,
-      "required_pages_per_second_upper_estimate": 0.36903950103451366
+      "required_token_positions_per_second": 28793.58046714918,
+      "required_pages_per_second_upper_estimate": 0.3693331340946009
     }
   ],
   "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": 0.11352874730468443,
-    "producer_p95_seconds": 0.242173989998264,
-    "gpu_update_seconds": 0.9609780440747908,
-    "producer_batches_per_second": 8.808341708521063,
-    "gpu_batches_per_second": 1.040606501018219,
-    "producer_headroom": 8.464622986596972,
+    "producer_batch_seconds": 0.06949543899995092,
+    "producer_p95_seconds": 0.23346730900084367,
+    "gpu_update_seconds": 0.9602140321362194,
+    "producer_batches_per_second": 14.389433528158678,
+    "gpu_batches_per_second": 1.0414344787018655,
+    "producer_headroom": 13.81693598822935,
     "sustainable": true,
-    "buffer_seconds": 60.54161677671182,
+    "buffer_seconds": 60.49348402458182,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 459517.8785198832,
-    "upload_bytes_per_second": 5402278.777431174,
+    "generation_bytes_per_second": 460172.93873254355,
+    "upload_bytes_per_second": 5347256.208164543,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.60183970162747,
+    "estimated_full_shards_per_hour": 98.74240037424306,
     "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": 1.4661580640004104,
-      "bytes": 8583866,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 1.0245029070010787,
       "bytes": 5870358,
@@ -562,6 +556,12 @@
       "bytes": 7643162,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 0.8460896450014843,
+      "bytes": 884554,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -218,3 +218,4 @@
 {"step": 578900, "loss": 0.581824004650116, "lr": 0.0003, "grad_norm": 0.16481607578175492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84733783930, "tokens_per_second": 152618.67735907383, "data_wait_s": 1.02440026239492e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 579000, "loss": 0.5822747945785522, "lr": 0.0003, "grad_norm": 0.16719512047122234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84748502214, "tokens_per_second": 153617.60331346476, "data_wait_s": 1.343400072073564e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 579100, "loss": 0.5804442167282104, "lr": 0.0003, "grad_norm": 0.16040496497450182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84763264848, "tokens_per_second": 153304.47354579673, "data_wait_s": 1.4106000890024006e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 579200, "loss": 0.5281848311424255, "lr": 0.0003, "grad_norm": 0.16683918222658417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84777790977, "tokens_per_second": 150649.3625108852, "data_wait_s": 1.4092001947574317e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0f48133a9729aa82163bafeb1c0d8b8d0dcb16427d26281fea00812b92002cfb
-size 469257
+oid sha256:438a610dcd564bc3bea53201bab5b5ec277f89598c4e4412ea4702a292b0a6ec
+size 471375