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": 75443.4144145719,
-      "required_pages_per_second_upper_estimate": 0.006539243951869382
+      "required_token_positions_per_second": 74892.89570167659,
+      "required_pages_per_second_upper_estimate": 0.006491526385112538
     },
     {
       "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": 9430.426801821488,
-      "required_pages_per_second_upper_estimate": 0.0037619382486921526
+      "required_token_positions_per_second": 9361.611962709574,
+      "required_pages_per_second_upper_estimate": 0.003734486980496878
     },
     {
       "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": 18860.853603642976,
-      "required_pages_per_second_upper_estimate": 0.004363126541516526
+      "required_token_positions_per_second": 18723.223925419148,
+      "required_pages_per_second_upper_estimate": 0.004331288337659052
     },
     {
       "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": 28291.280405464462,
-      "required_pages_per_second_upper_estimate": 0.0017227905438159954
+      "required_token_positions_per_second": 28084.835888128724,
+      "required_pages_per_second_upper_estimate": 0.0017102191558409227
     },
     {
       "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": 28291.280405464462,
-      "required_pages_per_second_upper_estimate": 0.0021779410808892173
+      "required_token_positions_per_second": 28084.835888128724,
+      "required_pages_per_second_upper_estimate": 0.0021620484104697122
     },
     {
       "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": 28291.280405464462,
-      "required_pages_per_second_upper_estimate": 0.35128301781124777
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28084.835888128724,
+      "required_pages_per_second_upper_estimate": 0.3487196678312915
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10324925296870902,
+    "producer_batch_seconds": 0.13491348410948945,
     "producer_p95_seconds": 0.3298561420015176,
-    "gpu_update_seconds": 0.9772622378257503,
-    "producer_batches_per_second": 9.685300099004712,
-    "gpu_batches_per_second": 1.0232667970726441,
-    "producer_headroom": 9.465078048767305,
+    "gpu_update_seconds": 0.9844458450863381,
+    "producer_batches_per_second": 7.412157551193674,
+    "gpu_batches_per_second": 1.0157999091481744,
+    "producer_headroom": 7.29686770439794,
     "sustainable": true,
-    "buffer_seconds": 60.590258745196515,
+    "buffer_seconds": 60.051196550266624,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318707.45691604743,
-    "upload_bytes_per_second": 3691317.598779273,
+    "generation_bytes_per_second": 329125.26561555656,
+    "upload_bytes_per_second": 3708029.676250744,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.38720112429684,
+    "estimated_full_shards_per_hour": 70.62262035703681,
     "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": 3.9525413480005227,
-      "bytes": 15333188,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 2.024789704999421,
       "bytes": 1170220,
@@ -562,6 +556,12 @@
       "bytes": 1239440,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 0.8729745059972629,
+      "bytes": 5859266,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -292,3 +292,4 @@
 {"step": 69200, "loss": 0.6179433986544609, "lr": 0.0003, "grad_norm": 0.13127947763161085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10127801907, "tokens_per_second": 150235.5464349099, "data_wait_s": 1.146899739978835e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 69300, "loss": 0.631734854541719, "lr": 0.0003, "grad_norm": 0.1502898218404227, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10142597199, "tokens_per_second": 151147.13258239228, "data_wait_s": 1.0574000043561682e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 69400, "loss": 0.5570238828659058, "lr": 0.0003, "grad_norm": 0.13479370695657739, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10157305576, "tokens_per_second": 150245.46984794692, "data_wait_s": 1.567399885971099e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 69500, "loss": 0.6252729836851358, "lr": 0.0003, "grad_norm": 0.13367581634287012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10171816427, "tokens_per_second": 148491.94474402617, "data_wait_s": 1.2938002328155562e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:012e58086cbdacc2f0a0f5ecbcf7714faff76a09e1aa9b5f9130fa2b5df3fd6f
-size 626772
+oid sha256:0a8cca6fb2376971a50105a688e8b2872a897150118658b935ceda2221f54582
+size 628890