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": 76003.3520565372,
-      "required_pages_per_second_upper_estimate": 0.006587777927525955
+      "required_token_positions_per_second": 75600.39146151379,
+      "required_pages_per_second_upper_estimate": 0.006552850324443617
     },
     {
       "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": 9500.41900706715,
-      "required_pages_per_second_upper_estimate": 0.0037898591858413714
+      "required_token_positions_per_second": 9450.048932689224,
+      "required_pages_per_second_upper_estimate": 0.003769765810072293
     },
     {
       "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": 19000.8380141343,
-      "required_pages_per_second_upper_estimate": 0.004395509471242213
+      "required_token_positions_per_second": 18900.097865378448,
+      "required_pages_per_second_upper_estimate": 0.0043722050107936105
     },
     {
       "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": 28501.25702120145,
-      "required_pages_per_second_upper_estimate": 0.0017355770180522163
+      "required_token_positions_per_second": 28350.146798067668,
+      "required_pages_per_second_upper_estimate": 0.0017263751982774377
     },
     {
       "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": 28501.25702120145,
-      "required_pages_per_second_upper_estimate": 0.002194105662021125
+      "required_token_positions_per_second": 28350.146798067668,
+      "required_pages_per_second_upper_estimate": 0.002182472778744416
     },
     {
       "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": 28501.25702120145,
-      "required_pages_per_second_upper_estimate": 0.3538902246321747
+      "required_token_positions_per_second": 28350.146798067668,
+      "required_pages_per_second_upper_estimate": 0.35201394139423703
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.08440637911721183,
-    "producer_p95_seconds": 0.22595442199963145,
-    "gpu_update_seconds": 0.970062477575402,
-    "producer_batches_per_second": 11.847445779084294,
-    "gpu_batches_per_second": 1.0308614373987792,
-    "producer_headroom": 11.492762605398749,
+    "producer_batch_seconds": 0.3825043607890706,
+    "producer_p95_seconds": 0.2990865149986348,
+    "gpu_update_seconds": 0.9752330454205786,
+    "producer_batches_per_second": 2.6143492794097667,
+    "gpu_batches_per_second": 1.025395934536591,
+    "producer_headroom": 2.5495998095518817,
     "sustainable": true,
-    "buffer_seconds": 60.14387360967492,
+    "buffer_seconds": 60.46444881607587,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 89,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333722.22536115145,
-    "upload_bytes_per_second": 5842737.837577686,
+    "generation_bytes_per_second": 309760.882472733,
+    "upload_bytes_per_second": 5852443.887126167,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.60902090669543,
+    "estimated_full_shards_per_hour": 66.46747451435559,
     "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": 1.4425641930001802,
-      "bytes": 7147715,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 1.1025827090006715,
       "bytes": 1108574,
@@ -562,6 +556,12 @@
       "bytes": 5709641,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.273396311000397,
+      "bytes": 6852766,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -146,3 +146,4 @@
 {"step": 129600, "loss": 0.6036495044827461, "lr": 0.0003, "grad_norm": 0.14175531830058616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18966413322, "tokens_per_second": 148983.96393344173, "data_wait_s": 8.854000043356791e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 129700, "loss": 0.5633120387792587, "lr": 0.0003, "grad_norm": 0.1401520968925492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18980861179, "tokens_per_second": 147496.29808510488, "data_wait_s": 1.1600001016631722e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 129800, "loss": 0.5880979876965284, "lr": 0.0003, "grad_norm": 0.1268623129422515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18995416582, "tokens_per_second": 142832.72843818384, "data_wait_s": 1.3171000318834558e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 129900, "loss": 0.603065631352365, "lr": 0.0003, "grad_norm": 0.1387297366005279, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19010098040, "tokens_per_second": 134676.9472382548, "data_wait_s": 7.50599974708166e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d48fd9aa8f5270ce58aca63d14f550321cdcf6cd4f349b237a0a1793b53e1db9
-size 315195
+oid sha256:18cef231e78ae047b36b828668c6dc42e08d0826b52a5f4ad5140e26ea9d5e10
+size 317313