Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75763.6191346008,
-      "required_pages_per_second_upper_estimate": 0.006566998485450043
+      "required_token_positions_per_second": 75622.34723820254,
+      "required_pages_per_second_upper_estimate": 0.006554753395520589
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9470.4523918251,
-      "required_pages_per_second_upper_estimate": 0.003777905054980493
+      "required_token_positions_per_second": 9452.793404775317,
+      "required_pages_per_second_upper_estimate": 0.00377086062102095
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18940.9047836502,
-      "required_pages_per_second_upper_estimate": 0.004381644973158547
+      "required_token_positions_per_second": 18905.586809550634,
+      "required_pages_per_second_upper_estimate": 0.0043734747814257275
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28411.3571754753,
-      "required_pages_per_second_upper_estimate": 0.0017301025891155316
+      "required_token_positions_per_second": 28358.38021432595,
+      "required_pages_per_second_upper_estimate": 0.0017268765701301703
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28411.3571754753,
-      "required_pages_per_second_upper_estimate": 0.002187184923038425
+      "required_token_positions_per_second": 28358.38021432595,
+      "required_pages_per_second_upper_estimate": 0.0021831066099195335
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28411.3571754753,
-      "required_pages_per_second_upper_estimate": 0.3527739694236848
+      "required_token_positions_per_second": 28358.38021432595,
+      "required_pages_per_second_upper_estimate": 0.35211617286869323
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.17069391810153434,
-    "producer_p95_seconds": 0.29812347700135433,
-    "gpu_update_seconds": 0.973131970755194,
-    "producer_batches_per_second": 5.858439545603302,
-    "gpu_batches_per_second": 1.0276098515435221,
-    "producer_headroom": 5.701034820563104,
+    "producer_batch_seconds": 0.09406006507816755,
+    "producer_p95_seconds": 0.3139965790032875,
+    "gpu_update_seconds": 0.974949901617897,
+    "producer_batches_per_second": 10.631504445260179,
+    "gpu_batches_per_second": 1.025693728816766,
+    "producer_headroom": 10.365184212956645,
     "sustainable": true,
-    "buffer_seconds": 60.33418218682203,
+    "buffer_seconds": 60.44689390030961,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 383814.43950580515,
-    "upload_bytes_per_second": 5257994.821109479,
+    "generation_bytes_per_second": 402630.2698180239,
+    "upload_bytes_per_second": 5319985.196651535,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.35764397507302,
+    "estimated_full_shards_per_hour": 86.39508314996279,
     "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": 1.1495631180005148,
-      "bytes": 5725671,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.1277136550015712,
       "bytes": 7537147,
@@ -562,6 +556,12 @@
       "bytes": 5609510,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.1360651520008105,
+      "bytes": 10573263,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -209,3 +209,4 @@
 {"step": 175900, "loss": 0.5911736041307449, "lr": 0.0003, "grad_norm": 0.14925685939116468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25743945074, "tokens_per_second": 149222.00557064585, "data_wait_s": 1.1783002264564857e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 176000, "loss": 0.5919118411839008, "lr": 0.0003, "grad_norm": 0.13506766904797246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25758535359, "tokens_per_second": 148414.89522368903, "data_wait_s": 1.5083998732734472e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 176100, "loss": 0.5708627104759216, "lr": 0.0003, "grad_norm": 0.13841311921039212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25773104273, "tokens_per_second": 148764.03594994082, "data_wait_s": 1.2873999367002398e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 176200, "loss": 0.60870729573071, "lr": 0.0003, "grad_norm": 0.1381184597474122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25787859964, "tokens_per_second": 150172.95685653726, "data_wait_s": 1.4061999536352232e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1d97fc5452b83c4ca392a754324b64b3a2e8b5deac06314b6abfc5234fe24e0b
-size 450195
+oid sha256:032c2c63cd5c89de5ba6f485db9c1f9f73eed0d13eb2c90ab699b0fcc16d4a1e
+size 452313