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": 76956.82440747012,
-      "required_pages_per_second_upper_estimate": 0.005324051676942786
+      "required_token_positions_per_second": 76780.02194900489,
+      "required_pages_per_second_upper_estimate": 0.0053118200726279385
     },
     {
       "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": 9619.603050933765,
-      "required_pages_per_second_upper_estimate": 0.00402752691727142
+      "required_token_positions_per_second": 9597.502743625611,
+      "required_pages_per_second_upper_estimate": 0.004018273980108392
     },
     {
       "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": 19239.20610186753,
-      "required_pages_per_second_upper_estimate": 0.004007019259034812
+      "required_token_positions_per_second": 19195.005487251223,
+      "required_pages_per_second_upper_estimate": 0.003997813436658829
     },
     {
       "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": 28858.809152801296,
-      "required_pages_per_second_upper_estimate": 0.001730041543860299
+      "required_token_positions_per_second": 28792.508230876836,
+      "required_pages_per_second_upper_estimate": 0.0017260669048265735
     },
     {
       "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": 28858.809152801296,
-      "required_pages_per_second_upper_estimate": 0.002243645475143899
+      "required_token_positions_per_second": 28792.508230876836,
+      "required_pages_per_second_upper_estimate": 0.0022384908700912075
     },
     {
       "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": 28858.809152801296,
-      "required_pages_per_second_upper_estimate": 0.37016981763704027
+      "required_token_positions_per_second": 28792.508230876836,
+      "required_pages_per_second_upper_estimate": 0.3693193805989769
     }
   ],
   "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": 0.13699971171109837,
-    "producer_p95_seconds": 0.34622759299963946,
-    "gpu_update_seconds": 0.9580436896619567,
-    "producer_batches_per_second": 7.299285432868468,
-    "gpu_batches_per_second": 1.0437937338252783,
-    "producer_headroom": 6.99303434800108,
+    "producer_batch_seconds": 0.11951544600009356,
+    "producer_p95_seconds": 0.36750896599551197,
+    "gpu_update_seconds": 0.9602497906156896,
+    "producer_batches_per_second": 8.367119342877382,
+    "gpu_batches_per_second": 1.0413956970079874,
+    "producer_headroom": 8.034524597054492,
     "sustainable": true,
-    "buffer_seconds": 60.356752448703276,
+    "buffer_seconds": 60.49573680878844,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299933.082467768,
-    "upload_bytes_per_second": 4209964.663156165,
+    "generation_bytes_per_second": 346636.13747842767,
+    "upload_bytes_per_second": 4181553.2178794052,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.35865741276531,
+    "estimated_full_shards_per_hour": 74.38004582657454,
     "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": 1.8211984559966368,
-      "bytes": 13136869,
-      "pages": 32,
-      "shards": 4
-    },
     {
       "seconds": 1.4420257410019985,
       "bytes": 5442386,
@@ -562,6 +556,12 @@
       "bytes": 5807963,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.2447194750056951,
+      "bytes": 7832230,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -401,3 +401,4 @@
 {"step": 840100, "loss": 0.48094847798347473, "lr": 3e-05, "grad_norm": 0.3027066853642255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122966749524, "tokens_per_second": 137449.3540139752, "data_wait_s": 9.551993571221828e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 840200, "loss": 0.5049573183059692, "lr": 3e-05, "grad_norm": 0.3118878406648667, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122981334289, "tokens_per_second": 151254.18466818944, "data_wait_s": 1.0846997611224651e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 840300, "loss": 0.5117914080619812, "lr": 3e-05, "grad_norm": 0.2940921969676939, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122995970860, "tokens_per_second": 151514.29915577834, "data_wait_s": 8.223003533203155e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 840400, "loss": 0.538476288318634, "lr": 3e-05, "grad_norm": 0.2918500214189647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123010790877, "tokens_per_second": 152886.55978331104, "data_wait_s": 1.164599962066859e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e7f0cce6a05feb47c6eeeb658d3a141e2c5a23c8863656e5552446d8c32a9107
-size 859983
+oid sha256:b2af1910829298f96f318c3e66de1b9612800d865915b6f6bc6a4486c5e4509a
+size 862101