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": 76648.5177045791,
-      "required_pages_per_second_upper_estimate": 0.005302722303866677
+      "required_token_positions_per_second": 77385.78071356808,
+      "required_pages_per_second_upper_estimate": 0.0053537278695144006
     },
     {
       "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": 9581.064713072388,
-      "required_pages_per_second_upper_estimate": 0.00401139171998087
+      "required_token_positions_per_second": 9673.22258919601,
+      "required_pages_per_second_upper_estimate": 0.0040499762982385374
     },
     {
       "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": 19162.129426144777,
-      "required_pages_per_second_upper_estimate": 0.003990966220130391
+      "required_token_positions_per_second": 19346.44517839202,
+      "required_pages_per_second_upper_estimate": 0.004029354330590235
     },
     {
       "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": 28743.194139217165,
-      "required_pages_per_second_upper_estimate": 0.0017231106003298344
+      "required_token_positions_per_second": 29019.667767588027,
+      "required_pages_per_second_upper_estimate": 0.0017396847721998804
     },
     {
       "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": 28743.194139217165,
-      "required_pages_per_second_upper_estimate": 0.0022346569163744336
+      "required_token_positions_per_second": 29019.667767588027,
+      "required_pages_per_second_upper_estimate": 0.002256151524901296
     },
     {
       "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": 28743.194139217165,
-      "required_pages_per_second_upper_estimate": 0.36868683238051286
+      "required_token_positions_per_second": 29019.667767588027,
+      "required_pages_per_second_upper_estimate": 0.3722331392310005
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.07583416910166818,
-    "producer_p95_seconds": 0.19814987999961886,
-    "gpu_update_seconds": 0.9618972709187221,
-    "producer_batches_per_second": 13.186667854952502,
-    "gpu_batches_per_second": 1.0396120565399727,
-    "producer_headroom": 12.68421982219045,
+    "producer_batch_seconds": 0.0893907867422854,
+    "producer_p95_seconds": 0.19242108300022664,
+    "gpu_update_seconds": 0.9527331677752686,
+    "producer_batches_per_second": 11.186835203531777,
+    "gpu_batches_per_second": 1.0496118260846365,
+    "producer_headroom": 10.658068940840723,
     "sustainable": true,
-    "buffer_seconds": 60.5995280678795,
+    "buffer_seconds": 60.02218956984192,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 262974.4194342877,
-    "upload_bytes_per_second": 5413503.412403364,
+    "generation_bytes_per_second": 263742.0735565503,
+    "upload_bytes_per_second": 5457112.558486102,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.42818867942308,
+    "estimated_full_shards_per_hour": 56.592909383987255,
     "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": [
@@ -526,6 +526,12 @@
       "bytes": 1116344,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.331704049000109,
+      "bytes": 10313766,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -106,3 +106,4 @@
 {"step": 770600, "loss": 0.5072928071022034, "lr": 3e-05, "grad_norm": 0.25135627223391616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112793444564, "tokens_per_second": 153555.77624449757, "data_wait_s": 1.302700002270285e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 770700, "loss": 0.6074251532554626, "lr": 3e-05, "grad_norm": 0.2844986677756334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112808199028, "tokens_per_second": 153507.88402762378, "data_wait_s": 1.2989999959245324e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 770800, "loss": 0.568845272064209, "lr": 3e-05, "grad_norm": 0.24240980720523625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112822968721, "tokens_per_second": 154027.62018610426, "data_wait_s": 8.985000022221357e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 770900, "loss": 0.5253165364265442, "lr": 3e-05, "grad_norm": 0.24298080124792695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112837598704, "tokens_per_second": 151614.51904652428, "data_wait_s": 1.4668999938294291e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:97cb1f81086fd04679c57c6e25164785af6589a8d55c7bb087f3bea2a22d5947
-size 230475
+oid sha256:8793173b05f1a2580311d3422fae13db32fa9088780f18492eafd867cf6f2e3e
+size 232593