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": 76617.15618784979,
-      "required_pages_per_second_upper_estimate": 0.005300552641370598
+      "required_token_positions_per_second": 76662.02483399928,
+      "required_pages_per_second_upper_estimate": 0.005303656758420822
     },
     {
       "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": 9577.144523481224,
-      "required_pages_per_second_upper_estimate": 0.0040097504184619165
+      "required_token_positions_per_second": 9582.75310424991,
+      "required_pages_per_second_upper_estimate": 0.004012098614109281
     },
     {
       "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": 19154.289046962447,
-      "required_pages_per_second_upper_estimate": 0.0039893332759114155
+      "required_token_positions_per_second": 19165.50620849982,
+      "required_pages_per_second_upper_estimate": 0.003991669514843208
     },
     {
       "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": 28731.43357044367,
-      "required_pages_per_second_upper_estimate": 0.0017224055721892122
+      "required_token_positions_per_second": 28748.25931274973,
+      "required_pages_per_second_upper_estimate": 0.0017234142497490395
     },
     {
       "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": 28731.43357044367,
-      "required_pages_per_second_upper_estimate": 0.0022337425838815698
+      "required_token_positions_per_second": 28748.25931274973,
+      "required_pages_per_second_upper_estimate": 0.002235050711337247
     },
     {
       "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": 28731.43357044367,
-      "required_pages_per_second_upper_estimate": 0.3685359804318014
+      "required_token_positions_per_second": 28748.25931274973,
+      "required_pages_per_second_upper_estimate": 0.368751802988029
     }
   ],
   "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": 0.20290958921870583,
-    "producer_p95_seconds": 0.2763582470033725,
-    "gpu_update_seconds": 0.9622910020209291,
-    "producer_batches_per_second": 4.92830330912627,
-    "gpu_batches_per_second": 1.0391866887457926,
-    "producer_headroom": 4.742461929602179,
+    "producer_batch_seconds": 0.2363843506093417,
+    "producer_p95_seconds": 0.27845966200038674,
+    "gpu_update_seconds": 0.9617277936455175,
+    "producer_batches_per_second": 4.230398490518691,
+    "gpu_batches_per_second": 1.0397952587076726,
+    "producer_headroom": 4.0684918065278675,
     "sustainable": true,
-    "buffer_seconds": 60.624333127318536,
+    "buffer_seconds": 60.588850999667606,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 165,
+    "upload_pages_per_commit": 157,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 676495.0768080958,
-    "upload_bytes_per_second": 5123755.694367059,
+    "generation_bytes_per_second": 638236.4271997388,
+    "upload_bytes_per_second": 5131684.160697608,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.58004774180486,
+    "estimated_full_shards_per_hour": 68.4753399467188,
     "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": 1.203305928996997,
-      "bytes": 7121753,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.1241114900003595,
       "bytes": 6180535,
@@ -562,6 +556,12 @@
       "bytes": 6128565,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.9614488920014992,
+      "bytes": 6537093,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -319,3 +319,4 @@
 {"step": 751900, "loss": 0.4962722361087799, "lr": 5e-05, "grad_norm": 0.24667855449304538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110056944555, "tokens_per_second": 149582.60326599845, "data_wait_s": 1.2260999938007444e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 752000, "loss": 0.5134796500205994, "lr": 5e-05, "grad_norm": 0.2307066487088763, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110071415461, "tokens_per_second": 149253.18554392314, "data_wait_s": 1.3985001714900136e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 752100, "loss": 0.5339847803115845, "lr": 5e-05, "grad_norm": 0.2317741747614533, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110085950757, "tokens_per_second": 150050.41510498594, "data_wait_s": 9.28099689190276e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 752200, "loss": 0.5571590662002563, "lr": 5e-05, "grad_norm": 0.2566676285867222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110100594628, "tokens_per_second": 151454.34595932436, "data_wait_s": 1.3125998520990834e-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:ad26530d80c0d1f2cf4f710b9dddaa669dafd6f2351b5df7cdd216772eaaa32b
-size 684741
+oid sha256:fdd77e7a31a93e07286d08dcf5ce1f9c7d15c0b92e0f2ded65f55618fb801bc4
+size 686859