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": 77741.3088934432,
-      "required_pages_per_second_upper_estimate": 0.005378324133937193
+      "required_token_positions_per_second": 77259.88965457807,
+      "required_pages_per_second_upper_estimate": 0.005345018433945971
     },
     {
       "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": 9717.6636116804,
-      "required_pages_per_second_upper_estimate": 0.004068582826318672
+      "required_token_positions_per_second": 9657.48620682226,
+      "required_pages_per_second_upper_estimate": 0.00404338780355168
     },
     {
       "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": 19435.3272233608,
-      "required_pages_per_second_upper_estimate": 0.0040478661165800804
+      "required_token_positions_per_second": 19314.97241364452,
+      "required_pages_per_second_upper_estimate": 0.004022799383685952
     },
     {
       "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": 29152.990835041204,
-      "required_pages_per_second_upper_estimate": 0.0017476772865211616
+      "required_token_positions_per_second": 28972.458620466776,
+      "required_pages_per_second_upper_estimate": 0.0017368546559141552
     },
     {
       "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": 29152.990835041204,
-      "required_pages_per_second_upper_estimate": 0.0022665168069695824
+      "required_token_positions_per_second": 28972.458620466776,
+      "required_pages_per_second_upper_estimate": 0.002252481221363712
     },
     {
       "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": 29152.990835041204,
-      "required_pages_per_second_upper_estimate": 0.37394326438913306
+      "required_token_positions_per_second": 28972.458620466776,
+      "required_pages_per_second_upper_estimate": 0.371627590980962
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,30 +155,37 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.029110802624987997,
-    "producer_p95_seconds": 0.073699526999917,
-    "gpu_update_seconds": 0.9483761085249531,
-    "producer_batches_per_second": 34.35150905601018,
-    "gpu_batches_per_second": 1.0544339856424045,
-    "producer_headroom": 32.578150480498614,
+    "producer_batch_seconds": 0.22783827374219712,
+    "producer_p95_seconds": 0.14410103000000163,
+    "gpu_update_seconds": 0.9542855979943949,
+    "producer_batches_per_second": 4.389078198211408,
+    "gpu_batches_per_second": 1.0479043193166513,
+    "producer_headroom": 4.188434113024335,
     "sustainable": true,
-    "buffer_seconds": 60.696070945597,
+    "buffer_seconds": 60.119992673646884,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 61,
+    "upload_pages_per_commit": 57,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "generation_bytes_per_second": 423180.14523553813,
+    "upload_bytes_per_second": 43101.512019429436,
+    "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 90.80460803794487,
     "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": [],
+  "upload_measurements": [
+    {
+      "seconds": 1.0425156310002421,
+      "bytes": 44934,
+      "pages": 1,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -3,3 +3,4 @@
 {"step": 485300, "loss": 0.5296840071678162, "lr": 0.0003, "grad_norm": 0.16333932344688673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71031548750, "tokens_per_second": 153642.53780209355, "data_wait_s": 7.868000011512777e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 485400, "loss": 0.5068361163139343, "lr": 0.0003, "grad_norm": 0.16174502470969496, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71046213893, "tokens_per_second": 154217.95321105205, "data_wait_s": 1.3796000075672055e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 485500, "loss": 0.5999634265899658, "lr": 0.0003, "grad_norm": 0.1477095802620134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71060987918, "tokens_per_second": 153889.3690089966, "data_wait_s": 1.7192000086652115e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 485600, "loss": 0.6141518950462341, "lr": 0.0003, "grad_norm": 0.16510797658642953, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71075706176, "tokens_per_second": 153904.7417917476, "data_wait_s": 1.053000005413196e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c7e39c08856e0a2be74956d90b85487d21d3b88e13bbf350ef0d96d6ce945944
-size 10755
+oid sha256:2ad09996409aa722ac34717c8f771aa38671189aba1da3a08c69765cd885a1d3
+size 12873