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": 75455.43652671404,
-      "required_pages_per_second_upper_estimate": 0.006540285998080098
+      "required_token_positions_per_second": 74123.01558118123,
+      "required_pages_per_second_upper_estimate": 0.006424795127511326
     },
     {
       "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": 9431.929565839255,
-      "required_pages_per_second_upper_estimate": 0.003762537723727164
+      "required_token_positions_per_second": 9265.376947647654,
+      "required_pages_per_second_upper_estimate": 0.0036960973941469817
     },
     {
       "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": 18863.85913167851,
-      "required_pages_per_second_upper_estimate": 0.004363821817532061
+      "required_token_positions_per_second": 18530.753895295307,
+      "required_pages_per_second_upper_estimate": 0.0042867637835467395
     },
     {
       "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": 28295.78869751776,
-      "required_pages_per_second_upper_estimate": 0.0017230650751488608
+      "required_token_positions_per_second": 27796.13084294296,
+      "required_pages_per_second_upper_estimate": 0.0016926385333074182
     },
     {
       "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": 28295.78869751776,
-      "required_pages_per_second_upper_estimate": 0.0021782881416912312
+      "required_token_positions_per_second": 27796.13084294296,
+      "required_pages_per_second_upper_estimate": 0.0021398230969045967
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28295.78869751776,
-      "required_pages_per_second_upper_estimate": 0.35133899571026683
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27796.13084294296,
+      "required_pages_per_second_upper_estimate": 0.34513491740371455
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1106447075313497,
-    "producer_p95_seconds": 0.3171666960006405,
-    "gpu_update_seconds": 0.9771065332568521,
-    "producer_batches_per_second": 9.037937939477706,
-    "gpu_batches_per_second": 1.023429857404433,
-    "producer_headroom": 8.831028207833638,
+    "producer_batch_seconds": 0.3570526882107572,
+    "producer_p95_seconds": 0.33631493900247733,
+    "gpu_update_seconds": 0.9946708107045564,
+    "producer_batches_per_second": 2.800707102952074,
+    "gpu_batches_per_second": 1.0053577417152402,
+    "producer_headroom": 2.7857816046393493,
     "sustainable": true,
-    "buffer_seconds": 60.580605061924835,
+    "buffer_seconds": 60.67491945297794,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 365524.6833232774,
-    "upload_bytes_per_second": 4133610.891818876,
+    "generation_bytes_per_second": 357287.25017217634,
+    "upload_bytes_per_second": 4127625.2895953944,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.43308806203595,
+    "estimated_full_shards_per_hour": 76.6655266654393,
     "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": 1.3517716829992423,
-      "bytes": 11962174,
-      "pages": 33,
-      "shards": 4
-    },
     {
       "seconds": 0.8873195009982737,
       "bytes": 1168035,
@@ -562,6 +556,12 @@
       "bytes": 6045719,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.362017162002303,
+      "bytes": 11390848,
+      "pages": 38,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -285,3 +285,4 @@
 {"step": 343500, "loss": 0.6186619400978088, "lr": 0.0003, "grad_norm": 0.17305364622911118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50279126614, "tokens_per_second": 148423.57161131405, "data_wait_s": 1.7221998859895393e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 343600, "loss": 0.546442705206573, "lr": 0.0003, "grad_norm": 0.15118177201015093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50293865040, "tokens_per_second": 150371.70046929838, "data_wait_s": 1.4756999007659033e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 343700, "loss": 0.6093347147107124, "lr": 0.0003, "grad_norm": 0.14082922580876914, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50308394946, "tokens_per_second": 146465.03296410033, "data_wait_s": 1.4327000826597214e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 343800, "loss": 0.5572164366021752, "lr": 0.0003, "grad_norm": 0.1561984124133296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50322878578, "tokens_per_second": 130001.91361225491, "data_wait_s": 1.4461998944170773e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a3a6495ed1cf5b0cb3c675601227ca678522d3cd02627afef4a4ef729224d18c
-size 611946
+oid sha256:b8789a90ab4000a34101a6ad2ef588e6e78d2effb17c5755b530e7ba54b8879a
+size 614064