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": 77453.78804114848,
-      "required_pages_per_second_upper_estimate": 0.005358432774235153
+      "required_token_positions_per_second": 77637.55444671959,
+      "required_pages_per_second_upper_estimate": 0.00537114616056937
     },
     {
       "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": 9681.72350514356,
-      "required_pages_per_second_upper_estimate": 0.004053535454226465
+      "required_token_positions_per_second": 9704.694305839948,
+      "required_pages_per_second_upper_estimate": 0.004063152848793178
     },
     {
       "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": 19363.44701028712,
-      "required_pages_per_second_upper_estimate": 0.0040328953638055255
+      "required_token_positions_per_second": 19409.388611679897,
+      "required_pages_per_second_upper_estimate": 0.004042463787814138
     },
     {
       "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": 29045.17051543068,
-      "required_pages_per_second_upper_estimate": 0.0017412136229926062
+      "required_token_positions_per_second": 29114.082917519845,
+      "required_pages_per_second_upper_estimate": 0.0017453448162747063
     },
     {
       "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": 29045.17051543068,
-      "required_pages_per_second_upper_estimate": 0.0022581342513713298
+      "required_token_positions_per_second": 29114.082917519845,
+      "required_pages_per_second_upper_estimate": 0.0022634918875201397
     },
     {
       "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": 29045.17051543068,
-      "required_pages_per_second_upper_estimate": 0.3725602610976088
+      "required_token_positions_per_second": 29114.082917519845,
+      "required_pages_per_second_upper_estimate": 0.37344419539923607
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10561181802347619,
-    "producer_p95_seconds": 0.19905209799981094,
-    "gpu_update_seconds": 0.9518966323613623,
-    "producer_batches_per_second": 9.468637305132958,
-    "gpu_batches_per_second": 1.0505342344990842,
-    "producer_headroom": 9.013163963807228,
+    "producer_batch_seconds": 0.07191191254695184,
+    "producer_p95_seconds": 0.18205067400049302,
+    "gpu_update_seconds": 0.949643513701831,
+    "producer_batches_per_second": 13.905901881654064,
+    "gpu_batches_per_second": 1.0530267258940915,
+    "producer_headroom": 13.205649524086867,
     "sustainable": true,
-    "buffer_seconds": 60.92138447112719,
+    "buffer_seconds": 60.77718487691718,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 249907.33574326447,
-    "upload_bytes_per_second": 5858009.841121344,
+    "generation_bytes_per_second": 238683.42961644812,
+    "upload_bytes_per_second": 5815267.430068867,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.62429670546961,
+    "estimated_full_shards_per_hour": 51.215907729817225,
     "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": [
@@ -538,6 +538,12 @@
       "bytes": 2311911,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.5313736190000782,
+      "bytes": 6009990,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -76,3 +76,4 @@
 {"step": 492600, "loss": 0.5887793302536011, "lr": 0.0003, "grad_norm": 0.15616353349922218, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72100060969, "tokens_per_second": 153020.56320353612, "data_wait_s": 1.0361998647567816e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 492700, "loss": 0.5844504833221436, "lr": 0.0003, "grad_norm": 0.1607671126032676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72114861262, "tokens_per_second": 154501.30251059355, "data_wait_s": 1.7562000721227378e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 492800, "loss": 0.5880060791969299, "lr": 0.0003, "grad_norm": 0.1524960584163102, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72129413368, "tokens_per_second": 151991.147364005, "data_wait_s": 9.154999133897945e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 492900, "loss": 0.5948826670646667, "lr": 0.0003, "grad_norm": 0.15656816627424516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72144115352, "tokens_per_second": 126557.60718155648, "data_wait_s": 1.3593000403488986e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a41b93bfd73bade9f309b7db1758b896a973b7f14c0cac467603174bacc8c364
-size 166152
+oid sha256:ee6c973c8ea0f17d0027cef4d5eaddb9b826524a21d9120f5eeabe39ecc33b28
+size 168270