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": 76959.76413271973,
-      "required_pages_per_second_upper_estimate": 0.005324255053956657
+      "required_token_positions_per_second": 77038.9342920407,
+      "required_pages_per_second_upper_estimate": 0.005329732229278561
     },
     {
       "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": 9619.970516589967,
-      "required_pages_per_second_upper_estimate": 0.0040276807674681165
+      "required_token_positions_per_second": 9629.866786505088,
+      "required_pages_per_second_upper_estimate": 0.004031824129023962
     },
     {
       "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": 19239.941033179934,
-      "required_pages_per_second_upper_estimate": 0.0040071723258457434
+      "required_token_positions_per_second": 19259.733573010177,
+      "required_pages_per_second_upper_estimate": 0.004011294589928008
     },
     {
       "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": 28859.9115497699,
-      "required_pages_per_second_upper_estimate": 0.001730107630875304
+      "required_token_positions_per_second": 28889.60035951526,
+      "required_pages_per_second_upper_estimate": 0.001731887429687351
     },
     {
       "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": 28859.9115497699,
-      "required_pages_per_second_upper_estimate": 0.0022437311816592763
+      "required_token_positions_per_second": 28889.60035951526,
+      "required_pages_per_second_upper_estimate": 0.002246039356029705
     },
     {
       "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": 28859.9115497699,
-      "required_pages_per_second_upper_estimate": 0.37018395800169185
+      "required_token_positions_per_second": 28889.60035951526,
+      "required_pages_per_second_upper_estimate": 0.37056477417574507
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.11120015461699495,
-    "producer_p95_seconds": 0.2807819230001769,
-    "gpu_update_seconds": 0.9580070941076892,
-    "producer_batches_per_second": 8.99279325145082,
-    "gpu_batches_per_second": 1.0438336064008211,
-    "producer_headroom": 8.61515973073364,
+    "producer_batch_seconds": 0.10105882418733358,
+    "producer_p95_seconds": 0.30770943100287695,
+    "gpu_update_seconds": 0.9570225844572363,
+    "producer_batches_per_second": 9.895226943728256,
+    "gpu_batches_per_second": 1.0449074204107083,
+    "producer_headroom": 9.469955663477695,
     "sustainable": true,
-    "buffer_seconds": 60.35444692878442,
+    "buffer_seconds": 60.29242282080589,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 316565.0346165753,
+    "generation_bytes_per_second": 317074.3504088751,
     "upload_bytes_per_second": 5882695.566735004,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.92748717186875,
+    "estimated_full_shards_per_hour": 68.03677448463144,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -293,3 +293,4 @@
 {"step": 426400, "loss": 0.5964219570159912, "lr": 0.0003, "grad_norm": 0.16855262090960713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62413150360, "tokens_per_second": 133586.97409603183, "data_wait_s": 1.0431998816784471e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 426500, "loss": 0.555823028087616, "lr": 0.0003, "grad_norm": 0.1637813321546856, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62427777916, "tokens_per_second": 150993.46444015187, "data_wait_s": 1.0121999366674572e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 426600, "loss": 0.592617392539978, "lr": 0.0003, "grad_norm": 0.1528376389051259, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62442468509, "tokens_per_second": 152493.71875544576, "data_wait_s": 9.315997886005789e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 426700, "loss": 0.6029167771339417, "lr": 0.0003, "grad_norm": 0.16569533429502378, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62457150272, "tokens_per_second": 151457.6455850499, "data_wait_s": 1.4354998711496592e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:534852e5ebbc45ec452366b86eca2e1a76872fd721d6a71c4f579d5c4fda6cef
-size 629673
+oid sha256:1a370feca878074d105d3f3d0fcfb56f655c7962bca343bf1f9efe0630223898
+size 631791