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": 76341.82116612568,
-      "required_pages_per_second_upper_estimate": 0.005281504325702455
+      "required_token_positions_per_second": 76624.04493650892,
+      "required_pages_per_second_upper_estimate": 0.005301029220986935
     },
     {
       "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": 9542.72764576571,
-      "required_pages_per_second_upper_estimate": 0.003995340790468565
+      "required_token_positions_per_second": 9578.005617063614,
+      "required_pages_per_second_upper_estimate": 0.004010110940363185
     },
     {
       "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": 19085.45529153142,
-      "required_pages_per_second_upper_estimate": 0.003974997019923335
+      "required_token_positions_per_second": 19156.01123412723,
+      "required_pages_per_second_upper_estimate": 0.0039896919620807094
     },
     {
       "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": 28628.182937297126,
-      "required_pages_per_second_upper_estimate": 0.001716215854386665
+      "required_token_positions_per_second": 28734.016851190845,
+      "required_pages_per_second_upper_estimate": 0.001722560435925564
     },
     {
       "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": 28628.182937297126,
-      "required_pages_per_second_upper_estimate": 0.002225715301305965
+      "required_token_positions_per_second": 28734.016851190845,
+      "required_pages_per_second_upper_estimate": 0.002233943422597011
     },
     {
       "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": 28628.182937297126,
-      "required_pages_per_second_upper_estimate": 0.36721159217169
+      "required_token_positions_per_second": 28734.016851190845,
+      "required_pages_per_second_upper_estimate": 0.36856911598351544
     }
   ],
   "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": 0.07586474524200071,
-    "producer_p95_seconds": 0.2643924750009319,
-    "gpu_update_seconds": 0.9657616084316643,
-    "producer_batches_per_second": 13.181353167536557,
-    "gpu_batches_per_second": 1.0354522185075639,
-    "producer_headroom": 12.730044836385918,
+    "producer_batch_seconds": 0.28664050898413507,
+    "producer_p95_seconds": 0.2854869149996375,
+    "gpu_update_seconds": 0.9622044889576294,
+    "producer_batches_per_second": 3.4886904281046607,
+    "gpu_batches_per_second": 1.0392801233792985,
+    "producer_headroom": 3.3568335905058184,
     "sustainable": true,
-    "buffer_seconds": 60.84298133119485,
+    "buffer_seconds": 60.61888280433065,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 496797.2045888876,
-    "upload_bytes_per_second": 4879857.799763692,
+    "generation_bytes_per_second": 468584.58681644284,
+    "upload_bytes_per_second": 4836792.947686039,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.60111525773974,
+    "estimated_full_shards_per_hour": 100.54734423990215,
     "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": 1.2537715590005973,
-      "bytes": 5914249,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.3754841779991693,
       "bytes": 2754086,
@@ -562,6 +556,12 @@
       "bytes": 7915290,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.0522420070010412,
+      "bytes": 1136593,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -230,3 +230,4 @@
 {"step": 908000, "loss": 0.49739935994148254, "lr": 3e-05, "grad_norm": 0.32429283499638284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132905040804, "tokens_per_second": 153394.06379499636, "data_wait_s": 2.1040999854449183e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 908100, "loss": 0.5233396887779236, "lr": 3e-05, "grad_norm": 0.31039775129288033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132919705531, "tokens_per_second": 151719.47321947245, "data_wait_s": 1.51169988384936e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 908200, "loss": 0.5213196873664856, "lr": 3e-05, "grad_norm": 0.3281043705806001, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132934596573, "tokens_per_second": 154829.57237913038, "data_wait_s": 1.3740998838329688e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 908300, "loss": 0.5920027494430542, "lr": 3e-05, "grad_norm": 0.32380916206789356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132949354105, "tokens_per_second": 153057.10831939772, "data_wait_s": 1.5220000932458788e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:042e4f9e261f47614ef996a5884e04ef430dcab476b9b015e0d79a453c618920
-size 494673
+oid sha256:1965c2407142040125656c301e9499ff7246c13246043b460e1ea372c5f8e5e1
+size 496791