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": 76413.12669596064,
-      "required_pages_per_second_upper_estimate": 0.005286437407708062
+      "required_token_positions_per_second": 74012.03462873618,
+      "required_pages_per_second_upper_estimate": 0.005120324287196301
     },
     {
       "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": 9551.64083699508,
-      "required_pages_per_second_upper_estimate": 0.003999072557507704
+      "required_token_positions_per_second": 9251.504328592022,
+      "required_pages_per_second_upper_estimate": 0.0038734116690023472
     },
     {
       "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": 19103.28167399016,
-      "required_pages_per_second_upper_estimate": 0.003978709785276171
+      "required_token_positions_per_second": 18503.008657184044,
+      "required_pages_per_second_upper_estimate": 0.0038536887461394493
     },
     {
       "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": 28654.92251098524,
-      "required_pages_per_second_upper_estimate": 0.0017178188510003026
+      "required_token_positions_per_second": 27754.512985776066,
+      "required_pages_per_second_upper_estimate": 0.0016638406748097515
     },
     {
       "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": 28654.92251098524,
-      "required_pages_per_second_upper_estimate": 0.002227794185545261
+      "required_token_positions_per_second": 27754.512985776066,
+      "required_pages_per_second_upper_estimate": 0.0021577913054431925
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28654.92251098524,
-      "required_pages_per_second_upper_estimate": 0.36755457871224384
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 27754.512985776066,
+      "required_pages_per_second_upper_estimate": 0.35600509210728526
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08683022129690698,
-    "producer_p95_seconds": 0.30689670300125726,
-    "gpu_update_seconds": 0.9648604001424459,
-    "producer_batches_per_second": 11.516727529469298,
-    "gpu_batches_per_second": 1.036419361653112,
-    "producer_headroom": 11.11203433241527,
+    "producer_batch_seconds": 0.44045441935165286,
+    "producer_p95_seconds": 0.30885835099979886,
+    "gpu_update_seconds": 0.9961623183288911,
+    "producer_batches_per_second": 2.2703824869597087,
+    "gpu_batches_per_second": 1.0038524662100718,
+    "producer_headroom": 2.261669481703097,
     "sustainable": true,
-    "buffer_seconds": 60.78620520897409,
+    "buffer_seconds": 60.765901418062356,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 327857.06613796443,
-    "upload_bytes_per_second": 5745341.256823239,
+    "generation_bytes_per_second": 300848.00260945864,
+    "upload_bytes_per_second": 5739515.486149182,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.35049427131844,
+    "estimated_full_shards_per_hour": 64.55497797692126,
     "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": 0.9312373259999731,
-      "bytes": 6794347,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.2384252730007574,
       "bytes": 10759915,
@@ -562,6 +556,12 @@
       "bytes": 6262944,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.1364409210000304,
+      "bytes": 7535674,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -311,3 +311,4 @@
 {"step": 428200, "loss": 0.6211259961128235, "lr": 0.0003, "grad_norm": 0.2876302383040058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62676626534, "tokens_per_second": 153195.33449153855, "data_wait_s": 1.528899883851409e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 428300, "loss": 0.5747002959251404, "lr": 0.0003, "grad_norm": 0.1377744653043305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62691255767, "tokens_per_second": 151241.10010620445, "data_wait_s": 1.4224999176803976e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 428400, "loss": 0.5906609296798706, "lr": 0.0003, "grad_norm": 0.16108718480436882, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62706002550, "tokens_per_second": 153828.52408089308, "data_wait_s": 9.537998266750947e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 428500, "loss": 0.6180595755577087, "lr": 0.0003, "grad_norm": 0.16318784764526642, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62720771923, "tokens_per_second": 152615.36355919624, "data_wait_s": 1.1893000191776082e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ebbe6eda94236f7c4a130a602bbd4df2e2a02631e4db81720835e0bfe0dcc74c
-size 667797
+oid sha256:a60934b8085b1ca00683ec75d30aa119be54a17b48624cb5c6d5e4a21a213ca2
+size 669915