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": 77125.03292351439,
-      "required_pages_per_second_upper_estimate": 0.005335688732380259
+      "required_token_positions_per_second": 77055.1805246714,
+      "required_pages_per_second_upper_estimate": 0.005330856181348405
     },
     {
       "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": 9640.629115439298,
-      "required_pages_per_second_upper_estimate": 0.004036330091405731
+      "required_token_positions_per_second": 9631.897565583926,
+      "required_pages_per_second_upper_estimate": 0.004032674373816781
     },
     {
       "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": 19281.258230878597,
-      "required_pages_per_second_upper_estimate": 0.004015777608518599
+      "required_token_positions_per_second": 19263.79513116785,
+      "required_pages_per_second_upper_estimate": 0.00401214050538176
     },
     {
       "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": 28921.887346317893,
-      "required_pages_per_second_upper_estimate": 0.0017338229852467942
+      "required_token_positions_per_second": 28895.69269675178,
+      "required_pages_per_second_upper_estimate": 0.0017322526560022197
     },
     {
       "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": 28921.887346317893,
-      "required_pages_per_second_upper_estimate": 0.002248549527238163
+      "required_token_positions_per_second": 28895.69269675178,
+      "required_pages_per_second_upper_estimate": 0.0022465130084524847
     },
     {
       "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": 28921.887346317893,
-      "required_pages_per_second_upper_estimate": 0.37097891697538377
+      "required_token_positions_per_second": 28895.69269675178,
+      "required_pages_per_second_upper_estimate": 0.37064292013637307
     }
   ],
   "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": 0.3345458898281777,
-    "producer_p95_seconds": 0.2386816969992651,
-    "gpu_update_seconds": 0.9559542110421755,
-    "producer_batches_per_second": 2.989126545579737,
-    "gpu_batches_per_second": 1.0460752078384654,
-    "producer_headroom": 2.857468108584901,
+    "producer_batch_seconds": 0.09062450984370685,
+    "producer_p95_seconds": 0.2540982050013554,
+    "gpu_update_seconds": 0.9568208068293849,
+    "producer_batches_per_second": 11.034542440280488,
+    "gpu_batches_per_second": 1.0451277740433949,
+    "producer_headroom": 10.558079800702265,
     "sustainable": true,
-    "buffer_seconds": 60.225115295657055,
+    "buffer_seconds": 60.27971083025125,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326613.45672318863,
-    "upload_bytes_per_second": 5223459.441432747,
+    "generation_bytes_per_second": 325745.2501341733,
+    "upload_bytes_per_second": 5223106.882505676,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.08364464065308,
+    "estimated_full_shards_per_hour": 69.8973477174654,
     "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": 1.0105871389987442,
-      "bytes": 6118542,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3655256990005,
       "bytes": 10029580,
@@ -562,6 +556,12 @@
       "bytes": 6909103,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.019442289001745,
+      "bytes": 6135652,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -207,3 +207,4 @@
 {"step": 625100, "loss": 0.5289347767829895, "lr": 0.0001, "grad_norm": 0.176023465211531, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91497866150, "tokens_per_second": 137716.377216595, "data_wait_s": 1.5683002857258543e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 625200, "loss": 0.559563398361206, "lr": 0.0001, "grad_norm": 0.156229464540777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91512442589, "tokens_per_second": 151267.92498369116, "data_wait_s": 1.3631000911118463e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 625300, "loss": 0.5687322616577148, "lr": 0.0001, "grad_norm": 0.1808495224529712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91527167819, "tokens_per_second": 152496.99468937732, "data_wait_s": 2.0665000192821026e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 625400, "loss": 0.5248603224754333, "lr": 0.0001, "grad_norm": 0.1754910851709876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91541938960, "tokens_per_second": 153515.0365525391, "data_wait_s": 1.5395002264995128e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6f266ec508ffcca74c31af04701799c8ec2060ea1978f468befe69893ceab345
-size 446742
+oid sha256:25f9abbb8a3056528144f43101c2e4482bcd57e45d7c943701e42b5aa7bbc168
+size 448860