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": 76962.90984951225,
-      "required_pages_per_second_upper_estimate": 0.005324472681943432
+      "required_token_positions_per_second": 76787.81086186219,
+      "required_pages_per_second_upper_estimate": 0.0053123589276921745
     },
     {
       "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": 9620.36373118903,
-      "required_pages_per_second_upper_estimate": 0.00402784539820949
+      "required_token_positions_per_second": 9598.476357732774,
+      "required_pages_per_second_upper_estimate": 0.004018681612003687
     },
     {
       "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": 19240.72746237806,
-      "required_pages_per_second_upper_estimate": 0.004007336118308181
+      "required_token_positions_per_second": 19196.952715465548,
+      "required_pages_per_second_upper_estimate": 0.003998218992944012
     },
     {
       "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": 28861.091193567092,
-      "required_pages_per_second_upper_estimate": 0.0017301783487197338
+      "required_token_positions_per_second": 28795.42907319832,
+      "required_pages_per_second_upper_estimate": 0.0017262420048638786
     },
     {
       "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": 28861.091193567092,
-      "required_pages_per_second_upper_estimate": 0.0022438228937758033
+      "required_token_positions_per_second": 28795.42907319832,
+      "required_pages_per_second_upper_estimate": 0.0022387179527342755
     },
     {
       "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": 28861.091193567092,
-      "required_pages_per_second_upper_estimate": 0.3701990892057194
+      "required_token_positions_per_second": 28795.42907319832,
+      "required_pages_per_second_upper_estimate": 0.36935684602812074
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07015329667186165,
-    "producer_p95_seconds": 0.16928101499979675,
-    "gpu_update_seconds": 0.9579679373371205,
-    "producer_batches_per_second": 14.254497613668068,
-    "gpu_batches_per_second": 1.0438762729154765,
-    "producer_headroom": 13.655351676742505,
+    "producer_batch_seconds": 0.08325434225783823,
+    "producer_p95_seconds": 0.16521600900068734,
+    "gpu_update_seconds": 0.9601523884127046,
+    "producer_batches_per_second": 12.011385507112717,
+    "gpu_batches_per_second": 1.0415013408998235,
+    "producer_headroom": 11.53276048280002,
     "sustainable": true,
-    "buffer_seconds": 60.35198005223859,
+    "buffer_seconds": 60.48960047000039,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 246546.48366005224,
-    "upload_bytes_per_second": 3882021.5945386295,
+    "generation_bytes_per_second": 231892.6525213204,
+    "upload_bytes_per_second": 4250137.562067343,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.90313608504463,
+    "estimated_full_shards_per_hour": 49.75876504640301,
     "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": [
@@ -244,6 +244,12 @@
       "bytes": 1228924,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.3698066919996563,
+      "bytes": 11420874,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -52,3 +52,4 @@
 {"step": 725200, "loss": 0.5401269793510437, "lr": 5e-05, "grad_norm": 0.22418393062260913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106149120929, "tokens_per_second": 150576.0619204012, "data_wait_s": 1.4080000255489722e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 725300, "loss": 0.5074473023414612, "lr": 5e-05, "grad_norm": 0.22292254839442308, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106163763190, "tokens_per_second": 151666.87422990502, "data_wait_s": 1.1142999937874265e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 725400, "loss": 0.535947322845459, "lr": 5e-05, "grad_norm": 0.21267017077298314, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106178423732, "tokens_per_second": 150817.89433524545, "data_wait_s": 1.3257999853522051e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 725500, "loss": 0.5536834597587585, "lr": 5e-05, "grad_norm": 0.2220945458908904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106193169137, "tokens_per_second": 152365.59023065472, "data_wait_s": 1.809900004445808e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:47ba1f193b0758e002c5275c3c925392d0cb9186865cc23ecac97b5710174bb7
-size 115320
+oid sha256:660c974db7d90e48d272e557cb40c1931f20e7dd17cc74084754309791a89ba1
+size 117438