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": 77117.96323719306,
-      "required_pages_per_second_upper_estimate": 0.005335199635076617
+      "required_token_positions_per_second": 76460.871917921,
+      "required_pages_per_second_upper_estimate": 0.005289740532947457
     },
     {
       "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": 9639.745404649133,
-      "required_pages_per_second_upper_estimate": 0.0040359601001518685
+      "required_token_positions_per_second": 9557.608989740125,
+      "required_pages_per_second_upper_estimate": 0.004001571298432853
     },
     {
       "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": 19279.490809298266,
-      "required_pages_per_second_upper_estimate": 0.004015409501213457
+      "required_token_positions_per_second": 19115.21797948025,
+      "required_pages_per_second_upper_estimate": 0.0039811958029282125
     },
     {
       "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": 28919.2362139474,
-      "required_pages_per_second_upper_estimate": 0.0017336640539092278
+      "required_token_positions_per_second": 28672.826969220376,
+      "required_pages_per_second_upper_estimate": 0.0017188921959329741
     },
     {
       "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": 28919.2362139474,
-      "required_pages_per_second_upper_estimate": 0.002248343413357456
+      "required_token_positions_per_second": 28672.826969220376,
+      "required_pages_per_second_upper_estimate": 0.002229186178419654
     },
     {
       "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": 28919.2362139474,
-      "required_pages_per_second_upper_estimate": 0.3709449110958992
+      "required_token_positions_per_second": 28672.826969220376,
+      "required_pages_per_second_upper_estimate": 0.36778423787817466
     }
   ],
   "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": 0.24353179239824385,
-    "producer_p95_seconds": 0.33640491199912503,
-    "gpu_update_seconds": 0.9560418468681999,
-    "producer_batches_per_second": 4.106240052488569,
-    "gpu_batches_per_second": 1.045979319080852,
-    "producer_headroom": 3.9257373234653454,
+    "producer_batch_seconds": 0.11357934050789709,
+    "producer_p95_seconds": 0.3198495489996276,
+    "gpu_update_seconds": 0.964257902776015,
+    "producer_batches_per_second": 8.804418088080647,
+    "gpu_batches_per_second": 1.0370669476714545,
+    "producer_headroom": 8.489729720775856,
     "sustainable": true,
-    "buffer_seconds": 60.23063635269659,
+    "buffer_seconds": 60.748247874888946,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 439971.65977598616,
-    "upload_bytes_per_second": 5718201.65809837,
+    "generation_bytes_per_second": 440181.925760905,
+    "upload_bytes_per_second": 5773821.331752207,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.40767617187203,
+    "estimated_full_shards_per_hour": 94.452794357494,
     "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": 1.2862951650022296,
-      "bytes": 2384330,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.0045538510021288,
       "bytes": 5798532,
@@ -562,6 +556,12 @@
       "bytes": 12268120,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.4351412589967367,
+      "bytes": 7373378,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -364,3 +364,4 @@
 {"step": 721400, "loss": 0.5490963459014893, "lr": 8e-05, "grad_norm": 0.20821286446828496, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105592944425, "tokens_per_second": 150433.45233805044, "data_wait_s": 1.0651994671206921e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 721500, "loss": 0.5709020495414734, "lr": 8e-05, "grad_norm": 0.21111324364175713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105607461004, "tokens_per_second": 151306.48656521892, "data_wait_s": 1.5398996765725315e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 721600, "loss": 0.5116252899169922, "lr": 8e-05, "grad_norm": 0.20982151353438813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105622247717, "tokens_per_second": 153905.98184956878, "data_wait_s": 2.0448998839128762e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 721700, "loss": 0.5118348598480225, "lr": 8e-05, "grad_norm": 0.21285526642905558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105636903287, "tokens_per_second": 152914.39340622717, "data_wait_s": 1.3473996659740806e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d88e903e5ab6cc3f40222f228e687f5a4e33b4520bc2bcf7729a16c2ec3446ff
-size 780834
+oid sha256:519fa218303c4dfd92914f37cc804172b514741fd75a378d0e2274002c8db7e9
+size 782952