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": 75833.88118876625,
-      "required_pages_per_second_upper_estimate": 0.0065730886222275554
+      "required_token_positions_per_second": 75782.1544973889,
+      "required_pages_per_second_upper_estimate": 0.006568605083719068
     },
     {
       "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": 9479.23514859578,
-      "required_pages_per_second_upper_estimate": 0.003781408627970233
+      "required_token_positions_per_second": 9472.769312173612,
+      "required_pages_per_second_upper_estimate": 0.0037788293091485605
     },
     {
       "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": 18958.47029719156,
-      "required_pages_per_second_upper_estimate": 0.004385708445573862
+      "required_token_positions_per_second": 18945.538624347224,
+      "required_pages_per_second_upper_estimate": 0.004382716930650994
     },
     {
       "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": 28437.705445787342,
-      "required_pages_per_second_upper_estimate": 0.0017317070605388444
+      "required_token_positions_per_second": 28418.307936520836,
+      "required_pages_per_second_upper_estimate": 0.00173052585399538
     },
     {
       "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": 28437.705445787342,
-      "required_pages_per_second_upper_estimate": 0.0021892132858237264
+      "required_token_positions_per_second": 28418.307936520836,
+      "required_pages_per_second_upper_estimate": 0.002187720011864649
     },
     {
       "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": 28437.705445787342,
-      "required_pages_per_second_upper_estimate": 0.35310112675897215
+      "required_token_positions_per_second": 28418.307936520836,
+      "required_pages_per_second_upper_estimate": 0.35286027461316954
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.053492692531243335,
-    "producer_p95_seconds": 0.16403136899953097,
-    "gpu_update_seconds": 0.9722303387911233,
-    "producer_batches_per_second": 18.694142184298027,
-    "gpu_batches_per_second": 1.0285628416445074,
-    "producer_headroom": 18.1750121892495,
+    "producer_batch_seconds": 0.10078019728907606,
+    "producer_p95_seconds": 0.16359976500007178,
+    "gpu_update_seconds": 0.9728939549025402,
+    "producer_batches_per_second": 9.922584266546119,
+    "gpu_batches_per_second": 1.0278612534910603,
+    "producer_headroom": 9.653622249933775,
     "sustainable": true,
-    "buffer_seconds": 60.27828100504964,
+    "buffer_seconds": 60.31942520395749,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329075.05434625334,
-    "upload_bytes_per_second": 5586239.863034566,
+    "generation_bytes_per_second": 346496.52109647705,
+    "upload_bytes_per_second": 5589485.727561776,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.61184618750286,
+    "estimated_full_shards_per_hour": 74.35008740111097,
     "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": [
@@ -256,6 +256,12 @@
       "bytes": 10569380,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 1.112220934999641,
+      "bytes": 6262678,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -40,3 +40,4 @@
 {"step": 359000, "loss": 0.564917329698801, "lr": 0.0003, "grad_norm": 0.1610518875786081, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52547888928, "tokens_per_second": 129261.42408392014, "data_wait_s": 1.4450999515247531e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 359100, "loss": 0.6039392333477736, "lr": 0.0003, "grad_norm": 0.1673209109689572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52562505296, "tokens_per_second": 149229.92241035664, "data_wait_s": 1.8160000763600692e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 359200, "loss": 0.6014014109969139, "lr": 0.0003, "grad_norm": 0.156046113746063, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52577275485, "tokens_per_second": 151141.48779721805, "data_wait_s": 1.648499983275542e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 359300, "loss": 0.5911583099514246, "lr": 0.0003, "grad_norm": 0.15084249697825547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52591963907, "tokens_per_second": 149539.5361958171, "data_wait_s": 1.127700033975998e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:39bda8a40f57c31880092a012b00040c1cf0a2bb60cf5b8ed5067644f189df04
-size 89121
+oid sha256:88dcaf5dd844e288efc36236a89f428f07e15c8b8e151af5c6cac8093ca22a0d
+size 91239