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": 76950.59117427071,
-      "required_pages_per_second_upper_estimate": 0.005323620447406957
+      "required_token_positions_per_second": 75708.56283077114,
+      "required_pages_per_second_upper_estimate": 0.005237694044700852
     },
     {
       "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": 9618.823896783839,
-      "required_pages_per_second_upper_estimate": 0.0040272007016994345
+      "required_token_positions_per_second": 9463.570353846393,
+      "required_pages_per_second_upper_estimate": 0.00396219928533417
     },
     {
       "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": 19237.647793567678,
-      "required_pages_per_second_upper_estimate": 0.004006694704511306
+      "required_token_positions_per_second": 18927.140707692786,
+      "required_pages_per_second_upper_estimate": 0.003942024267146082
     },
     {
       "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": 28856.47169035152,
-      "required_pages_per_second_upper_estimate": 0.0017299014165555332
+      "required_token_positions_per_second": 28390.71106153918,
+      "required_pages_per_second_upper_estimate": 0.0017019797780335353
     },
     {
       "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": 28856.47169035152,
-      "required_pages_per_second_upper_estimate": 0.0022434637477198394
+      "required_token_positions_per_second": 28390.71106153918,
+      "required_pages_per_second_upper_estimate": 0.002207252907494177
     },
     {
       "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": 28856.47169035152,
-      "required_pages_per_second_upper_estimate": 0.3701398351785061
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28390.71106153918,
+      "required_pages_per_second_upper_estimate": 0.36416555792690164
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1493097121249889,
-    "producer_p95_seconds": 0.3504691619964433,
-    "gpu_update_seconds": 0.9581212941305093,
-    "producer_batches_per_second": 6.697487964901361,
-    "gpu_batches_per_second": 1.0437091901892186,
-    "producer_headroom": 6.417005836354804,
+    "producer_batch_seconds": 0.11380941329684902,
+    "producer_p95_seconds": 0.3411537049978506,
+    "gpu_update_seconds": 0.9738396456527879,
+    "producer_batches_per_second": 8.786619410748571,
+    "gpu_batches_per_second": 1.026863102630902,
+    "producer_headroom": 8.556758333449295,
     "sustainable": true,
-    "buffer_seconds": 60.36164153022209,
+    "buffer_seconds": 60.37805803047285,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 161,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 517970.9048809694,
-    "upload_bytes_per_second": 5414242.803828149,
+    "generation_bytes_per_second": 560263.0611541586,
+    "upload_bytes_per_second": 5370375.615186227,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.14449844190418,
+    "estimated_full_shards_per_hour": 60.10970533355984,
     "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": 1.1853109039984702,
-      "bytes": 6001312,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.160805725998216,
       "bytes": 12260097,
@@ -562,6 +556,12 @@
       "bytes": 11022627,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 0.9499405899987323,
+      "bytes": 1172384,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -361,3 +361,4 @@
 {"step": 433200, "loss": 0.5947546362876892, "lr": 0.0003, "grad_norm": 0.15544372888172547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63408219132, "tokens_per_second": 149199.0484397043, "data_wait_s": 1.79109993041493e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 433300, "loss": 0.5636115670204163, "lr": 0.0003, "grad_norm": 0.15217058713028453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63422673884, "tokens_per_second": 150000.67706828352, "data_wait_s": 1.6509999113623053e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 433400, "loss": 0.5433777570724487, "lr": 0.0003, "grad_norm": 0.1578129144079159, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63437157888, "tokens_per_second": 149739.93214019344, "data_wait_s": 1.3639000826515257e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 433500, "loss": 0.6011660099029541, "lr": 0.0003, "grad_norm": 0.1556185541153759, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63451741509, "tokens_per_second": 151413.7340214323, "data_wait_s": 8.963994332589209e-06, "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 sha256:71395d63c6943c8729d507c2e35b2025b1d5d2ca86e381ec9fadc88a29a59c66
-size 774480
+oid sha256:d1e312009907f3475e2f9f83b2370fb5dee6465dfeed930c84801125111c35b0
+size 776598