Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75470.49210977585,
-      "required_pages_per_second_upper_estimate": 0.006541590977861869
+      "required_token_positions_per_second": 75436.64672980674,
+      "required_pages_per_second_upper_estimate": 0.006538657346106478
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9433.811513721981,
-      "required_pages_per_second_upper_estimate": 0.0037632884608752117
+      "required_token_positions_per_second": 9429.580841225843,
+      "required_pages_per_second_upper_estimate": 0.0037616007823623116
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18867.623027443962,
-      "required_pages_per_second_upper_estimate": 0.004364692528575095
+      "required_token_positions_per_second": 18859.161682451686,
+      "required_pages_per_second_upper_estimate": 0.004362735145325717
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28301.434541165945,
-      "required_pages_per_second_upper_estimate": 0.001723408877405583
+      "required_token_positions_per_second": 28288.742523677527,
+      "required_pages_per_second_upper_estimate": 0.0017226360001304077
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28301.434541165945,
-      "required_pages_per_second_upper_estimate": 0.002178722774363964
+      "required_token_positions_per_second": 28288.742523677527,
+      "required_pages_per_second_upper_estimate": 0.002177745707723953
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28301.434541165945,
-      "required_pages_per_second_upper_estimate": 0.35140909819295413
+      "required_token_positions_per_second": 28288.742523677527,
+      "required_pages_per_second_upper_estimate": 0.351251505813198
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3479336520934453,
-    "producer_p95_seconds": 0.38557227500132285,
-    "gpu_update_seconds": 0.9769116106035017,
-    "producer_batches_per_second": 2.8741111817819447,
-    "gpu_batches_per_second": 1.0236340618187914,
-    "producer_headroom": 2.807752583648133,
+    "producer_batch_seconds": 0.12062076641399244,
+    "producer_p95_seconds": 0.3791499630024191,
+    "gpu_update_seconds": 0.9773499114306254,
+    "producer_batches_per_second": 8.290446410925776,
+    "gpu_batches_per_second": 1.0231750044732901,
+    "producer_headroom": 8.102667065438654,
     "sustainable": true,
-    "buffer_seconds": 60.56851985741711,
+    "buffer_seconds": 60.59569450869878,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334578.18781686976,
-    "upload_bytes_per_second": 5958498.938318603,
+    "generation_bytes_per_second": 321789.67671194527,
+    "upload_bytes_per_second": 6059554.992129908,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.79269052390642,
+    "estimated_full_shards_per_hour": 69.04857374209183,
     "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": 0.8457645430025877,
-      "bytes": 857794,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.0460859910017462,
       "bytes": 6362637,
@@ -562,6 +556,12 @@
       "bytes": 8376549,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.3019169199978933,
+      "bytes": 10756780,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -328,3 +328,4 @@
 {"step": 117800, "loss": 0.5871414011344314, "lr": 0.0003, "grad_norm": 0.13882546866497125, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17238468392, "tokens_per_second": 149223.67364011874, "data_wait_s": 1.260400313185528e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 117900, "loss": 0.623930312693119, "lr": 0.0003, "grad_norm": 0.144380699745971, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17253039451, "tokens_per_second": 148450.04961165026, "data_wait_s": 2.0261999452486634e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 118000, "loss": 0.5838187895715237, "lr": 0.0003, "grad_norm": 0.14346968421489228, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17267660900, "tokens_per_second": 148297.7775362238, "data_wait_s": 1.4903002011124045e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 118100, "loss": 0.6580418385565281, "lr": 0.0003, "grad_norm": 0.15876329008956036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17282311564, "tokens_per_second": 148539.76533340095, "data_wait_s": 2.014799974858761e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:420cfd675593fe195bbe0939f78583cbec4b2e01574ca4c3ee9d5743dc600542
-size 703803
+oid sha256:6fedabdb3a99f66067c168cd5287222b7b8bd2684db94689f16bdf7d5907adba
+size 705921