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": 77161.9318077423,
-      "required_pages_per_second_upper_estimate": 0.005338241482808352
+      "required_token_positions_per_second": 76849.05879269981,
+      "required_pages_per_second_upper_estimate": 0.0053165962016622995
     },
     {
       "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": 9645.241475967787,
-      "required_pages_per_second_upper_estimate": 0.004038261190441969
+      "required_token_positions_per_second": 9606.132349087477,
+      "required_pages_per_second_upper_estimate": 0.004021887015708621
     },
     {
       "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": 19290.482951935574,
-      "required_pages_per_second_upper_estimate": 0.004017698874642514
+      "required_token_positions_per_second": 19212.264698174953,
+      "required_pages_per_second_upper_estimate": 0.0040014080751382426
     },
     {
       "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": 28935.724427903362,
-      "required_pages_per_second_upper_estimate": 0.001734652497159843
+      "required_token_positions_per_second": 28818.39704726243,
+      "required_pages_per_second_upper_estimate": 0.0017276188998389564
     },
     {
       "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": 28935.724427903362,
-      "required_pages_per_second_upper_estimate": 0.0022496252994685436
+      "required_token_positions_per_second": 28818.39704726243,
+      "required_pages_per_second_upper_estimate": 0.002240503611692319
     },
     {
       "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": 28935.724427903362,
-      "required_pages_per_second_upper_estimate": 0.3711564042008615
+      "required_token_positions_per_second": 28818.39704726243,
+      "required_pages_per_second_upper_estimate": 0.369651454538326
     }
   ],
   "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": 0.09427168898434957,
-    "producer_p95_seconds": 0.2587642529979348,
-    "gpu_update_seconds": 0.9554970731383666,
-    "producer_batches_per_second": 10.607638526196492,
-    "gpu_batches_per_second": 1.046575680986088,
-    "producer_headroom": 10.135567564690525,
+    "producer_batch_seconds": 0.13306676030472886,
+    "producer_p95_seconds": 0.27915533800114645,
+    "gpu_update_seconds": 0.9593871565672799,
+    "producer_batches_per_second": 7.515024771851025,
+    "gpu_batches_per_second": 1.0423320691284155,
+    "producer_headroom": 7.209818247398826,
     "sustainable": true,
-    "buffer_seconds": 60.196315607717096,
+    "buffer_seconds": 60.44139086373863,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 266683.71238171466,
-    "upload_bytes_per_second": 4981072.300653072,
+    "generation_bytes_per_second": 298700.3076699495,
+    "upload_bytes_per_second": 4981806.521031889,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.22411659802036,
+    "estimated_full_shards_per_hour": 64.09413263868201,
     "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": 1.084104961999401,
-      "bytes": 947822,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.1121349749992078,
       "bytes": 6275461,
@@ -562,6 +556,12 @@
       "bytes": 11492771,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.4007872010006395,
+      "bytes": 2588923,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -206,3 +206,4 @@
 {"step": 780600, "loss": 0.5281774401664734, "lr": 3e-05, "grad_norm": 0.23989931728272904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114257523595, "tokens_per_second": 151486.07673143584, "data_wait_s": 1.1620999430306256e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 780700, "loss": 0.5520626902580261, "lr": 3e-05, "grad_norm": 0.2511360282036276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114272261265, "tokens_per_second": 152739.50269140434, "data_wait_s": 1.0971998563036323e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 780800, "loss": 0.5353292226791382, "lr": 3e-05, "grad_norm": 0.2553284753201415, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114286941560, "tokens_per_second": 151826.8623804724, "data_wait_s": 1.5786001313244924e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 780900, "loss": 0.562856137752533, "lr": 3e-05, "grad_norm": 0.25590063527910917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114301683998, "tokens_per_second": 153222.26951529537, "data_wait_s": 2.138299896614626e-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 sha256:7c561d5fa204b3d7fc5a0a2d76036c111182896e2c7a072596fee8eb3bc2796c
-size 443841
+oid sha256:80c067450808da4b0fe52934366ca7b116d873207c4edf4dd4444265eae11d5c
+size 445959