Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76905.14364040564,
-      "required_pages_per_second_upper_estimate": 0.005320476281561365
+      "required_token_positions_per_second": 76241.68068399456,
+      "required_pages_per_second_upper_estimate": 0.005274576374790694
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9613.142955050705,
-      "required_pages_per_second_upper_estimate": 0.004024822210027325
+      "required_token_positions_per_second": 9530.21008549932,
+      "required_pages_per_second_upper_estimate": 0.00399009994938141
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19226.28591010141,
-      "required_pages_per_second_upper_estimate": 0.004004328323818299
+      "required_token_positions_per_second": 19060.42017099864,
+      "required_pages_per_second_upper_estimate": 0.003969782864536883
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28839.428865152113,
-      "required_pages_per_second_upper_estimate": 0.0017288797252077162
+      "required_token_positions_per_second": 28590.630256497963,
+      "required_pages_per_second_upper_estimate": 0.0017139646285123788
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28839.428865152113,
-      "required_pages_per_second_upper_estimate": 0.0022421387430240514
+      "required_token_positions_per_second": 28590.630256497963,
+      "required_pages_per_second_upper_estimate": 0.0022227957455506165
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28839.428865152113,
-      "required_pages_per_second_upper_estimate": 0.3699212281160082
+      "required_token_positions_per_second": 28590.630256497963,
+      "required_pages_per_second_upper_estimate": 0.36672990670332556
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13296261607854376,
-    "producer_p95_seconds": 0.3148557239983347,
-    "gpu_update_seconds": 0.9586875013814241,
-    "producer_batches_per_second": 7.520910986057008,
-    "gpu_batches_per_second": 1.0430927685601894,
-    "producer_headroom": 7.210203361335096,
+    "producer_batch_seconds": 0.20412410848427953,
+    "producer_p95_seconds": 0.34014576399931684,
+    "gpu_update_seconds": 0.9670300987406976,
+    "producer_batches_per_second": 4.898980367510163,
+    "gpu_batches_per_second": 1.034093976291159,
+    "producer_headroom": 4.737461468522091,
     "sustainable": true,
-    "buffer_seconds": 60.39731258702972,
+    "buffer_seconds": 60.922896220663944,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 165,
+    "upload_pages_per_commit": 166,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 739001.5516179414,
-    "upload_bytes_per_second": 3816169.204784275,
+    "generation_bytes_per_second": 681359.2384486482,
+    "upload_bytes_per_second": 3858716.2048604926,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.2862649507698,
+    "estimated_full_shards_per_hour": 73.10191566989224,
     "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": 1.7052124439978797,
-      "bytes": 6177071,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 2.175897373002954,
       "bytes": 7190309,
@@ -562,6 +556,12 @@
       "bytes": 7534957,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.5817108970004483,
+      "bytes": 10450177,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -343,3 +343,4 @@
 {"step": 834300, "loss": 0.5587754845619202, "lr": 3e-05, "grad_norm": 0.2948067504570078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122118046312, "tokens_per_second": 153036.59667053417, "data_wait_s": 1.3212003977969289e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 834400, "loss": 0.5076045393943787, "lr": 3e-05, "grad_norm": 0.2948557973666099, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122132533332, "tokens_per_second": 150243.17243182645, "data_wait_s": 1.3503005902748555e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 834500, "loss": 0.5453035235404968, "lr": 3e-05, "grad_norm": 0.3164018617622983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122147057444, "tokens_per_second": 149809.1086840898, "data_wait_s": 9.306997526437044e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 834600, "loss": 0.5199332237243652, "lr": 3e-05, "grad_norm": 0.27532710486133033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122161579239, "tokens_per_second": 150612.0426535751, "data_wait_s": 1.3378004950936884e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4ca16c31e6b049df81b3403c450408e2f0c27d5ca105e0fd59aae95ee8afa9ce
-size 735573
+oid sha256:6984440acfc55f5d18860b5a8ce887e79851efa62f4b768c4f66dd8e73c8c695
+size 737691