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": 76036.73247388365,
-      "required_pages_per_second_upper_estimate": 0.005260397582069855
+      "required_token_positions_per_second": 76859.6318403662,
+      "required_pages_per_second_upper_estimate": 0.005317327669632711
     },
     {
       "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": 9504.591559235456,
-      "required_pages_per_second_upper_estimate": 0.0039793740074103925
+      "required_token_positions_per_second": 9607.453980045775,
+      "required_pages_per_second_upper_estimate": 0.004022440354992068
     },
     {
       "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": 19009.18311847091,
-      "required_pages_per_second_upper_estimate": 0.0039591115377072975
+      "required_token_positions_per_second": 19214.90796009155,
+      "required_pages_per_second_upper_estimate": 0.004001958596887965
     },
     {
       "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": 28513.77467770637,
-      "required_pages_per_second_upper_estimate": 0.0017093572539154968
+      "required_token_positions_per_second": 28822.361940137325,
+      "required_pages_per_second_upper_estimate": 0.0017278565891127654
     },
     {
       "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": 28513.77467770637,
-      "required_pages_per_second_upper_estimate": 0.0022168205623515347
+      "required_token_positions_per_second": 28822.361940137325,
+      "required_pages_per_second_upper_estimate": 0.0022408118646736214
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28513.77467770637,
-      "required_pages_per_second_upper_estimate": 0.36574408585967816
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28822.361940137325,
+      "required_pages_per_second_upper_estimate": 0.36970231192695485
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.36272893781259086,
-    "producer_p95_seconds": 0.1925137790003646,
-    "gpu_update_seconds": 0.9696366164251385,
-    "producer_batches_per_second": 2.7568795752288846,
-    "gpu_batches_per_second": 1.0313141882850974,
-    "producer_headroom": 2.6731713832165087,
+    "producer_batch_seconds": 0.15239325770306777,
+    "producer_p95_seconds": 0.2016909369995119,
+    "gpu_update_seconds": 0.9592551803153253,
+    "producer_batches_per_second": 6.561970096790374,
+    "gpu_batches_per_second": 1.0424754752653835,
+    "producer_headroom": 6.2946038084204226,
     "sustainable": true,
-    "buffer_seconds": 60.11747021835859,
+    "buffer_seconds": 60.433076359865495,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317064.07755217253,
+    "generation_bytes_per_second": 316832.0659552692,
     "upload_bytes_per_second": 4025891.498281773,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.03457016872294,
+    "estimated_full_shards_per_hour": 67.98478588098104,
     "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": [

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

@@ -89,3 +89,4 @@
 {"step": 808900, "loss": 0.5370275974273682, "lr": 3e-05, "grad_norm": 0.27748202659556104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118399350596, "tokens_per_second": 150115.66276547694, "data_wait_s": 1.4887000361341052e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 809000, "loss": 0.5002292990684509, "lr": 3e-05, "grad_norm": 0.2541201568625616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118413899670, "tokens_per_second": 143029.28804722716, "data_wait_s": 1.671800055191852e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 809100, "loss": 0.5107089877128601, "lr": 3e-05, "grad_norm": 0.26920593157008044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118428649495, "tokens_per_second": 152458.50065118438, "data_wait_s": 1.3953000234323554e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 809200, "loss": 0.5278820395469666, "lr": 3e-05, "grad_norm": 0.2742863426345246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118443379648, "tokens_per_second": 152686.85749597132, "data_wait_s": 9.901999874273315e-06, "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 sha256:174794285b85387e66b5bb1e9f409680c8d6e434eba4349f128f0d142b13921a
-size 193686
+oid sha256:b631bd4226897da8419fb3d48131eaea86c5f06117497126118737ee3224fab9
+size 195804