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": 76676.2374048575,
-      "required_pages_per_second_upper_estimate": 0.005304640017050502
+      "required_token_positions_per_second": 76757.8175968003,
+      "required_pages_per_second_upper_estimate": 0.00531028392402123
     },
     {
       "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": 9584.529675607188,
-      "required_pages_per_second_upper_estimate": 0.0040128424274375445
+      "required_token_positions_per_second": 9594.727199600038,
+      "required_pages_per_second_upper_estimate": 0.004017111917784835
     },
     {
       "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": 19169.059351214375,
-      "required_pages_per_second_upper_estimate": 0.003992409540767974
+      "required_token_positions_per_second": 19189.454399200076,
+      "required_pages_per_second_upper_estimate": 0.003996657291409812
     },
     {
       "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": 28753.58902682156,
-      "required_pages_per_second_upper_estimate": 0.0017237337579696435
+      "required_token_positions_per_second": 28784.181598800114,
+      "required_pages_per_second_upper_estimate": 0.0017255677359476053
     },
     {
       "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": 28753.58902682156,
-      "required_pages_per_second_upper_estimate": 0.002235465073163397
+      "required_token_positions_per_second": 28784.181598800114,
+      "required_pages_per_second_upper_estimate": 0.0022378435110722264
     },
     {
       "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": 28753.58902682156,
-      "required_pages_per_second_upper_estimate": 0.3688201668375413
+      "required_token_positions_per_second": 28784.181598800114,
+      "required_pages_per_second_upper_estimate": 0.36921257550313763
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06332463564844204,
-    "producer_p95_seconds": 0.16122264200021164,
-    "gpu_update_seconds": 0.961549529493857,
-    "producer_batches_per_second": 15.791642379936896,
-    "gpu_batches_per_second": 1.039988029037238,
-    "producer_headroom": 15.184446300363573,
+    "producer_batch_seconds": 0.23980248547649552,
+    "producer_p95_seconds": 0.16978790699977253,
+    "gpu_update_seconds": 0.960527569807735,
+    "producer_batches_per_second": 4.17009856262735,
+    "gpu_batches_per_second": 1.0410945312066013,
+    "producer_headroom": 4.005494638219178,
     "sustainable": true,
-    "buffer_seconds": 60.57762035811299,
+    "buffer_seconds": 60.5132368978873,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 508147.34133439994,
-    "upload_bytes_per_second": 3666967.333548445,
+    "generation_bytes_per_second": 502156.14731350273,
+    "upload_bytes_per_second": 3694194.463902225,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.03659038566587,
+    "estimated_full_shards_per_hour": 107.7510196166402,
     "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": [
@@ -370,6 +370,12 @@
       "bytes": 11764874,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 0.9162838960000954,
+      "bytes": 5018838,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -41,3 +41,4 @@
 {"step": 444300, "loss": 0.5492525696754456, "lr": 0.0003, "grad_norm": 0.20942501199811075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65032080794, "tokens_per_second": 153140.43664036572, "data_wait_s": 1.4805000319029205e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 444400, "loss": 0.5989384055137634, "lr": 0.0003, "grad_norm": 0.16974348781045961, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65046721194, "tokens_per_second": 152128.5663943477, "data_wait_s": 1.2406000678311102e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 444500, "loss": 0.58826744556427, "lr": 0.0003, "grad_norm": 0.17988697609140758, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65061485024, "tokens_per_second": 152900.03011147288, "data_wait_s": 1.3471999409375712e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 444600, "loss": 0.5913227796554565, "lr": 0.0003, "grad_norm": 0.1416641029051212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65076195746, "tokens_per_second": 152711.02291267272, "data_wait_s": 1.5451999388460536e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fb5bc81e0b0546cfe2bd0c712d361e2fd34f2f83a13bbbd30b867007ee8602ab
-size 91239
+oid sha256:6b64901c26afc13f07eb194c50be7d2b7d82eb428af2c5e0cc610dc288f6cc42
+size 93357