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": 76791.06646873527,
-      "required_pages_per_second_upper_estimate": 0.0053125841581297024
+      "required_token_positions_per_second": 76832.51914823147,
+      "required_pages_per_second_upper_estimate": 0.005315451950680717
     },
     {
       "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": 9598.883308591909,
-      "required_pages_per_second_upper_estimate": 0.004018851993830307
+      "required_token_positions_per_second": 9604.064893528934,
+      "required_pages_per_second_upper_estimate": 0.004021021415239641
     },
     {
       "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": 19197.766617183817,
-      "required_pages_per_second_upper_estimate": 0.003998388507207896
+      "required_token_positions_per_second": 19208.129787057867,
+      "required_pages_per_second_upper_estimate": 0.004000546882197493
     },
     {
       "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": 28796.649925775724,
-      "required_pages_per_second_upper_estimate": 0.00172631519311176
+      "required_token_positions_per_second": 28812.194680586803,
+      "required_pages_per_second_upper_estimate": 0.001727247077427218
     },
     {
       "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": 28796.649925775724,
-      "required_pages_per_second_upper_estimate": 0.0022388128686522055
+      "required_token_positions_per_second": 28812.194680586803,
+      "required_pages_per_second_upper_estimate": 0.0022400214049646166
     },
     {
       "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": 28796.649925775724,
-      "required_pages_per_second_upper_estimate": 0.3693725058141343
+      "required_token_positions_per_second": 28812.194680586803,
+      "required_pages_per_second_upper_estimate": 0.3695718972381935
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.2341921430547984,
-    "producer_p95_seconds": 0.22175879000133136,
-    "gpu_update_seconds": 0.9601116821319006,
-    "producer_batches_per_second": 4.269998074897034,
-    "gpu_batches_per_second": 1.0415454978940872,
-    "producer_headroom": 4.099675034389368,
+    "producer_batch_seconds": 0.1067881651406708,
+    "producer_p95_seconds": 0.23122435900040728,
+    "gpu_update_seconds": 0.959593682692585,
+    "producer_batches_per_second": 9.364333572757914,
+    "gpu_batches_per_second": 1.0421077358429833,
+    "producer_headroom": 8.98595533904458,
     "sustainable": true,
-    "buffer_seconds": 60.487035974309734,
+    "buffer_seconds": 60.45440200963286,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 397296.77897722455,
-    "upload_bytes_per_second": 5590781.050795965,
+    "generation_bytes_per_second": 413376.8845352367,
+    "upload_bytes_per_second": 5523075.256719317,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.25064017283967,
+    "estimated_full_shards_per_hour": 88.70105649988962,
     "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": 0.8920637669998541,
-      "bytes": 2452797,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.1147250890003306,
       "bytes": 15773840,
@@ -562,6 +556,12 @@
       "bytes": 6406862,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 3.478892785999051,
+      "bytes": 11471341,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -113,3 +113,4 @@
 {"step": 568400, "loss": 0.5631968975067139, "lr": 0.0003, "grad_norm": 0.13816458641959578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83196871354, "tokens_per_second": 152554.49916727495, "data_wait_s": 1.3218999811215326e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 568500, "loss": 0.6140328645706177, "lr": 0.0003, "grad_norm": 0.1451410438802144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83211602486, "tokens_per_second": 153911.20691224738, "data_wait_s": 1.0392999683972448e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 568600, "loss": 0.57574862241745, "lr": 0.0003, "grad_norm": 0.1586142382442521, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83226314899, "tokens_per_second": 153039.4483900401, "data_wait_s": 1.5919000361463986e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 568700, "loss": 0.5584443807601929, "lr": 0.0003, "grad_norm": 0.16764500432431012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83240882348, "tokens_per_second": 151947.46953504998, "data_wait_s": 1.3409000530373305e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe97c42cca652abe5ab221a289414e8b6709c4b374b65ff184813d4594511635
-size 245301
+oid sha256:0ea35994ca9e8713b43c301253d694a20377c2ec43570499ca0c1db5ebd7c61b
+size 247419