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": 75199.94318443316,
-      "required_pages_per_second_upper_estimate": 0.006518140482713147
+      "required_token_positions_per_second": 75272.36870187912,
+      "required_pages_per_second_upper_estimate": 0.006524418143004567
     },
     {
       "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": 9399.992898054144,
-      "required_pages_per_second_upper_estimate": 0.0037497977094519486
+      "required_token_positions_per_second": 9409.04608773489,
+      "required_pages_per_second_upper_estimate": 0.003753409162172846
     },
     {
       "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": 18799.98579610829,
-      "required_pages_per_second_upper_estimate": 0.004349045845480202
+      "required_token_positions_per_second": 18818.09217546978,
+      "required_pages_per_second_upper_estimate": 0.00435323443768409
     },
     {
       "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": 28199.978694162437,
-      "required_pages_per_second_upper_estimate": 0.0017172307486207604
+      "required_token_positions_per_second": 28227.138263204673,
+      "required_pages_per_second_upper_estimate": 0.0017188846238801864
     },
     {
       "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": 28199.978694162437,
-      "required_pages_per_second_upper_estimate": 0.0021709124224138744
+      "required_token_positions_per_second": 28227.138263204673,
+      "required_pages_per_second_upper_estimate": 0.002173003241221245
     },
     {
       "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": 28199.978694162437,
-      "required_pages_per_second_upper_estimate": 0.3501493561240478
+      "required_token_positions_per_second": 28227.138263204673,
+      "required_pages_per_second_upper_estimate": 0.350486587074322
     }
   ],
   "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": 0.05418492963296728,
-    "producer_p95_seconds": 0.192757041000732,
-    "gpu_update_seconds": 0.9804262726525854,
-    "producer_batches_per_second": 18.45531602188477,
-    "gpu_batches_per_second": 1.0199645071673336,
-    "producer_headroom": 18.094076697962027,
+    "producer_batch_seconds": 0.09794492713278657,
+    "producer_p95_seconds": 0.20372904199939512,
+    "gpu_update_seconds": 0.9794829267563547,
+    "producer_batches_per_second": 10.209819224677895,
+    "gpu_batches_per_second": 1.0209468411170672,
+    "producer_headroom": 10.000343615840801,
     "sustainable": true,
-    "buffer_seconds": 60.786428904460294,
+    "buffer_seconds": 60.727941458893994,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 414109.6410558688,
-    "upload_bytes_per_second": 5692992.339418219,
+    "generation_bytes_per_second": 414852.7788829394,
+    "upload_bytes_per_second": 5707088.341010404,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.85828899151848,
+    "estimated_full_shards_per_hour": 89.0177490698446,
     "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": [
@@ -484,6 +484,12 @@
       "bytes": 6072045,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1727791039993463,
+      "bytes": 7506500,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -86,3 +86,4 @@
 {"step": 363600, "loss": 0.5906962864100933, "lr": 0.0003, "grad_norm": 0.1517345435929581, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53220693262, "tokens_per_second": 149365.92333611613, "data_wait_s": 1.32630011648871e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 363700, "loss": 0.6128485035151243, "lr": 0.0003, "grad_norm": 0.19333841950586778, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53235437534, "tokens_per_second": 150949.02513239713, "data_wait_s": 1.2414000593707897e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 363800, "loss": 0.5747907608747482, "lr": 0.0003, "grad_norm": 0.13842675587279343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53250257557, "tokens_per_second": 151172.31249218542, "data_wait_s": 1.116900057240855e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 363900, "loss": 0.6292346157133579, "lr": 0.0003, "grad_norm": 0.14552456501589275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53264934729, "tokens_per_second": 149934.00301102104, "data_wait_s": 8.941999112721533e-06, "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 sha256:8bb97bdfa90a4c4452706f01513c44ebd13a73e30755d4b95435fa9729dd45e8
-size 187332
+oid sha256:890a5bb9021b4db0addb3e147303913321681a03bebc7f21cafb8356eacec9bd
+size 189450