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": 75263.7261548585,
-      "required_pages_per_second_upper_estimate": 0.006523669028933154
+      "required_token_positions_per_second": 74978.09828105186,
+      "required_pages_per_second_upper_estimate": 0.006498911528748823
     },
     {
       "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": 9407.965769357312,
-      "required_pages_per_second_upper_estimate": 0.0037529782070198307
+      "required_token_positions_per_second": 9372.262285131483,
+      "required_pages_per_second_upper_estimate": 0.003738735553347488
     },
     {
       "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": 18815.931538714623,
-      "required_pages_per_second_upper_estimate": 0.004352734612396692
+      "required_token_positions_per_second": 18744.524570262965,
+      "required_pages_per_second_upper_estimate": 0.004336215866965124
     },
     {
       "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": 28223.897308071937,
-      "required_pages_per_second_upper_estimate": 0.0017186872667165802
+      "required_token_positions_per_second": 28116.786855394446,
+      "required_pages_per_second_upper_estimate": 0.0017121648021136355
     },
     {
       "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": 28223.897308071937,
-      "required_pages_per_second_upper_estimate": 0.002172753743523589
+      "required_token_positions_per_second": 28116.786855394446,
+      "required_pages_per_second_upper_estimate": 0.002164508084375766
     },
     {
       "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": 28223.897308071937,
-      "required_pages_per_second_upper_estimate": 0.3504463452583525
+      "required_token_positions_per_second": 28116.786855394446,
+      "required_pages_per_second_upper_estimate": 0.3491163919117231
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.06593840624221059,
-    "producer_p95_seconds": 0.2125187459987501,
-    "gpu_update_seconds": 0.9795954009545226,
-    "producer_batches_per_second": 15.165668340947073,
-    "gpu_batches_per_second": 1.020829619070889,
-    "producer_headroom": 14.856218959193358,
+    "producer_batch_seconds": 0.1324160578982827,
+    "producer_p95_seconds": 0.19453586199961137,
+    "gpu_update_seconds": 0.9833271540661658,
+    "producer_batches_per_second": 7.551954165318563,
+    "gpu_batches_per_second": 1.0169555430915236,
+    "producer_headroom": 7.426041597020829,
     "sustainable": true,
-    "buffer_seconds": 60.7349148591804,
+    "buffer_seconds": 60.966283552102276,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363395.696565661,
-    "upload_bytes_per_second": 3557442.5139041827,
+    "generation_bytes_per_second": 368650.8703346751,
+    "upload_bytes_per_second": 3556734.1987121,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.9762570641267,
+    "estimated_full_shards_per_hour": 79.10389502077284,
     "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": 1.1101808009998422,
-      "bytes": 6059135,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.0066997960002482,
       "bytes": 2574525,
@@ -562,6 +556,12 @@
       "bytes": 7622267,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 0.9423913660011749,
+      "bytes": 5379592,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -101,3 +101,4 @@
 {"step": 247200, "loss": 0.6252014748752117, "lr": 0.0003, "grad_norm": 0.15737314129582805, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36183114465, "tokens_per_second": 150194.20104432647, "data_wait_s": 1.375800093228463e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 247300, "loss": 0.585365790873766, "lr": 0.0003, "grad_norm": 0.14005982366468347, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36197816949, "tokens_per_second": 150224.37752116265, "data_wait_s": 1.0218998795608059e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 247400, "loss": 0.5953868869692087, "lr": 0.0003, "grad_norm": 0.1426853728589685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36212627121, "tokens_per_second": 150115.96048686322, "data_wait_s": 1.442299981135875e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 247500, "loss": 0.5994236934930086, "lr": 0.0003, "grad_norm": 0.13597782829188143, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36227420155, "tokens_per_second": 150848.63109803176, "data_wait_s": 1.4650999219156802e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1b300a8aaa333f5f31fba548a057156f1ada53b7648fd4e4f806595c17fc69b2
-size 219885
+oid sha256:cb8fcbd66658f7be155f5a7a42db4b3f22b0a24fe63a1ffe2a71df943e5fb87e
+size 222003