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": 75938.45905140482,
-      "required_pages_per_second_upper_estimate": 0.006582153166310898
+      "required_token_positions_per_second": 75191.62054073365,
+      "required_pages_per_second_upper_estimate": 0.006517419097050821
     },
     {
       "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": 9492.307381425602,
-      "required_pages_per_second_upper_estimate": 0.003786623337093347
+      "required_token_positions_per_second": 9398.952567591707,
+      "required_pages_per_second_upper_estimate": 0.0037493827060761553
     },
     {
       "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": 18984.614762851204,
-      "required_pages_per_second_upper_estimate": 0.00439175650757734
+      "required_token_positions_per_second": 18797.905135183413,
+      "required_pages_per_second_upper_estimate": 0.0043485645211936135
     },
     {
       "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": 28476.92214427681,
-      "required_pages_per_second_upper_estimate": 0.0017340951517227376
+      "required_token_positions_per_second": 28196.85770277512,
+      "required_pages_per_second_upper_estimate": 0.0017170406966198492
     },
     {
       "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": 28476.92214427681,
-      "required_pages_per_second_upper_estimate": 0.002192232295832215
+      "required_token_positions_per_second": 28196.85770277512,
+      "required_pages_per_second_upper_estimate": 0.002170672159857422
     },
     {
       "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": 28476.92214427681,
-      "required_pages_per_second_upper_estimate": 0.35358806690436456
+      "required_token_positions_per_second": 28196.85770277512,
+      "required_pages_per_second_upper_estimate": 0.35011060385630355
     }
   ],
   "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": 0.08756026350775414,
-    "producer_p95_seconds": 0.22659823800131562,
-    "gpu_update_seconds": 0.9708914418462389,
-    "producer_batches_per_second": 11.42070569387268,
-    "gpu_batches_per_second": 1.0299812696859378,
-    "producer_headroom": 11.088265418025596,
+    "producer_batch_seconds": 0.10345025044523481,
+    "producer_p95_seconds": 0.30846759399719303,
+    "gpu_update_seconds": 0.9805347919062236,
+    "producer_batches_per_second": 9.66648215636159,
+    "gpu_batches_per_second": 1.0198516240876419,
+    "producer_headroom": 9.478322069653236,
     "sustainable": true,
-    "buffer_seconds": 60.19526939446681,
+    "buffer_seconds": 60.793157098185866,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 186559.715878379,
-    "upload_bytes_per_second": 6298586.757238512,
+    "generation_bytes_per_second": 191030.73881196335,
+    "upload_bytes_per_second": 6198985.636322501,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 40.0313721395829,
+    "estimated_full_shards_per_hour": 40.99074958104301,
     "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": 1.6571312910000415,
-      "bytes": 16761254,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 0.7977973630004271,
       "bytes": 2044807,
@@ -562,6 +556,12 @@
       "bytes": 6909652,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.018912650000857,
+      "bytes": 5819747,
+      "pages": 22,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -172,3 +172,4 @@
 {"step": 17200, "loss": 0.6627637622877955, "lr": 0.0003, "grad_norm": 0.14520324430035478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2516775760, "tokens_per_second": 149344.46905586615, "data_wait_s": 1.0747997293947265e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 17300, "loss": 0.6741339080035686, "lr": 0.0003, "grad_norm": 0.13986137077484798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2531337238, "tokens_per_second": 147541.16995837618, "data_wait_s": 9.084000339498743e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 17400, "loss": 0.6582036688923836, "lr": 0.0003, "grad_norm": 0.14631516266041225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2546013116, "tokens_per_second": 150291.4679919444, "data_wait_s": 1.1026000720448792e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 17500, "loss": 0.6692678108811378, "lr": 0.0003, "grad_norm": 0.12959394541071664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2560852096, "tokens_per_second": 131618.66746332165, "data_wait_s": 2.067500099656172e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3d81a7b09dc099a8f46bcdec4747fbb1fc51d1d864c652961408ec256898bc44
-size 365272
+oid sha256:4f383b33c56735d98028b883706baff48dcbd6b66444e9fbd77bbcee4a177f85
+size 367390