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": 77087.14916098374,
-      "required_pages_per_second_upper_estimate": 0.005333067845785949
+      "required_token_positions_per_second": 75858.07693841434,
+      "required_pages_per_second_upper_estimate": 0.005248037777588134
     },
     {
       "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": 9635.893645122967,
-      "required_pages_per_second_upper_estimate": 0.004034347448872148
+      "required_token_positions_per_second": 9482.259617301792,
+      "required_pages_per_second_upper_estimate": 0.0039700240896667445
     },
     {
       "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": 19271.787290245935,
-      "required_pages_per_second_upper_estimate": 0.004013805061350316
+      "required_token_positions_per_second": 18964.519234603584,
+      "required_pages_per_second_upper_estimate": 0.003949809228563558
     },
     {
       "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": 28907.680935368902,
-      "required_pages_per_second_upper_estimate": 0.0017329713326023357
+      "required_token_positions_per_second": 28446.778851905376,
+      "required_pages_per_second_upper_estimate": 0.00170534095645543
     },
     {
       "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": 28907.680935368902,
-      "required_pages_per_second_upper_estimate": 0.0022474450412742772
+      "required_token_positions_per_second": 28446.778851905376,
+      "required_pages_per_second_upper_estimate": 0.002211611931579002
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28907.680935368902,
-      "required_pages_per_second_upper_estimate": 0.37079669238938573
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28446.778851905376,
+      "required_pages_per_second_upper_estimate": 0.3648847353408162
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06671783441409218,
-    "producer_p95_seconds": 0.2299591219998547,
-    "gpu_update_seconds": 0.956424005848644,
-    "producer_batches_per_second": 14.988496086269542,
-    "gpu_batches_per_second": 1.045561376423933,
-    "producer_headroom": 14.335357468476637,
+    "producer_batch_seconds": 0.11376124721869019,
+    "producer_p95_seconds": 0.2325220589991659,
+    "gpu_update_seconds": 0.9719202354662425,
+    "producer_batches_per_second": 8.790339631892738,
+    "gpu_batches_per_second": 1.0288910175023644,
+    "producer_headroom": 8.543508964857434,
     "sustainable": true,
-    "buffer_seconds": 60.25471236846457,
+    "buffer_seconds": 60.259054598907035,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 465858.33011053683,
-    "upload_bytes_per_second": 5187608.429521802,
+    "generation_bytes_per_second": 414591.4623856491,
+    "upload_bytes_per_second": 5021328.824447816,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.96235301482275,
+    "estimated_full_shards_per_hour": 88.96167663266282,
     "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": 1.3326483809996716,
-      "bytes": 16444186,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 0.9647154859994771,
       "bytes": 2573191,
@@ -562,6 +556,12 @@
       "bytes": 6026132,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.9889250640007958,
+      "bytes": 952891,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -164,3 +164,4 @@
 {"step": 776400, "loss": 0.5531155467033386, "lr": 3e-05, "grad_norm": 0.2534336618671871, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113642821074, "tokens_per_second": 151769.46908746462, "data_wait_s": 1.591500040376559e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 776500, "loss": 0.4901812672615051, "lr": 3e-05, "grad_norm": 0.26054071581981747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113657478276, "tokens_per_second": 152388.68438564215, "data_wait_s": 9.945997589966282e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 776600, "loss": 0.5543330311775208, "lr": 3e-05, "grad_norm": 0.24975549134984354, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113672165451, "tokens_per_second": 151397.1042602939, "data_wait_s": 1.3367000065045431e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 776700, "loss": 0.5264233350753784, "lr": 3e-05, "grad_norm": 0.2669087977298909, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113686785137, "tokens_per_second": 151662.1950079904, "data_wait_s": 1.0474002920091152e-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:c4caf5555cad3aa6ca426967d6e084d4abf94d144b24f6117552e9446cb36fe4
-size 354102
+oid sha256:8dfe30536d4eceee7c229ccf49068ccf5e77aa544a0e8f40733f60bfeac44f03
+size 356220