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": 76935.8055811443,
-      "required_pages_per_second_upper_estimate": 0.00532259754576717
+      "required_token_positions_per_second": 77047.43453849011,
+      "required_pages_per_second_upper_estimate": 0.0053303202960512195
     },
     {
       "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": 9616.975697643038,
-      "required_pages_per_second_upper_estimate": 0.004026426899313968
+      "required_token_positions_per_second": 9630.929317311264,
+      "required_pages_per_second_upper_estimate": 0.0040322689884843416
     },
     {
       "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": 19233.951395286076,
-      "required_pages_per_second_upper_estimate": 0.004005924842229827
+      "required_token_positions_per_second": 19261.858634622527,
+      "required_pages_per_second_upper_estimate": 0.004011737184220217
     },
     {
       "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": 28850.927092929112,
-      "required_pages_per_second_upper_estimate": 0.00172956902640617
+      "required_token_positions_per_second": 28892.787951933795,
+      "required_pages_per_second_upper_estimate": 0.0017320785209856701
     },
     {
       "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": 28850.927092929112,
-      "required_pages_per_second_upper_estimate": 0.002243032679658354
+      "required_token_positions_per_second": 28892.787951933795,
+      "required_pages_per_second_upper_estimate": 0.0022462871773195085
     },
     {
       "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": 28850.927092929112,
-      "required_pages_per_second_upper_estimate": 0.3700687150360964
+      "required_token_positions_per_second": 28892.787951933795,
+      "required_pages_per_second_upper_estimate": 0.37060566118871996
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07903171873437032,
-    "producer_p95_seconds": 0.18429463600023155,
-    "gpu_update_seconds": 0.9583054267526838,
-    "producer_batches_per_second": 12.653147571812928,
-    "gpu_batches_per_second": 1.0435086477477253,
-    "producer_headroom": 12.125579983570873,
+    "producer_batch_seconds": 0.3790223053906132,
+    "producer_p95_seconds": 0.20991658000002644,
+    "gpu_update_seconds": 0.9569170010867546,
+    "producer_batches_per_second": 2.6383671508974094,
+    "gpu_batches_per_second": 1.0450227123818647,
+    "producer_headroom": 2.524698381802554,
     "sustainable": true,
-    "buffer_seconds": 60.373241885419084,
+    "buffer_seconds": 60.285771068465536,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319208.6056802502,
+    "generation_bytes_per_second": 295331.38970850164,
     "upload_bytes_per_second": 5220488.687869276,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.49473598294858,
+    "estimated_full_shards_per_hour": 63.3712412685517,
     "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": [

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

@@ -64,3 +64,4 @@
 {"step": 891400, "loss": 0.5025626420974731, "lr": 3e-05, "grad_norm": 0.3141751969306708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130476014189, "tokens_per_second": 152513.9970211436, "data_wait_s": 9.995999789680354e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 891500, "loss": 0.49394935369491577, "lr": 3e-05, "grad_norm": 0.31786236779176874, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130490479344, "tokens_per_second": 151138.07208173414, "data_wait_s": 8.01799978944473e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 891600, "loss": 0.5052629113197327, "lr": 3e-05, "grad_norm": 0.31842241688430495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130505239684, "tokens_per_second": 154089.34495502705, "data_wait_s": 1.0707000001275446e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 891700, "loss": 0.5582869052886963, "lr": 3e-05, "grad_norm": 0.33334995990102567, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130519945362, "tokens_per_second": 146788.8785535297, "data_wait_s": 1.2253999557287898e-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:e42f0ab1d6f3099d20f5d3beddd939f60e52e9ad7dace4fcd38466df71b3aa25
-size 140736
+oid sha256:e08a4f8d45f91119ada5423c2016653d7e71ffdca5dc332a78aed449dcd9a4f4
+size 142854