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": 76302.29045879748,
-      "required_pages_per_second_upper_estimate": 0.005278769499645612
+      "required_token_positions_per_second": 75751.07596665632,
+      "required_pages_per_second_upper_estimate": 0.005240635202085455
     },
     {
       "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": 9537.786307349685,
-      "required_pages_per_second_upper_estimate": 0.003993271955260655
+      "required_token_positions_per_second": 9468.88449583204,
+      "required_pages_per_second_upper_estimate": 0.003964424205611657
     },
     {
       "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": 19075.57261469937,
-      "required_pages_per_second_upper_estimate": 0.003972938718962932
+      "required_token_positions_per_second": 18937.76899166408,
+      "required_pages_per_second_upper_estimate": 0.003944237858410606
     },
     {
       "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": 28613.358922049054,
-      "required_pages_per_second_upper_estimate": 0.001715327177307504
+      "required_token_positions_per_second": 28406.65348749612,
+      "required_pages_per_second_upper_estimate": 0.0017029355021269791
     },
     {
       "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": 28613.358922049054,
-      "required_pages_per_second_upper_estimate": 0.0022245627993243795
+      "required_token_positions_per_second": 28406.65348749612,
+      "required_pages_per_second_upper_estimate": 0.002208492360989008
     },
     {
       "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": 28613.358922049054,
-      "required_pages_per_second_upper_estimate": 0.36702144562087524
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28406.65348749612,
+      "required_pages_per_second_upper_estimate": 0.364370050249434
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.07323090928124998,
-    "producer_p95_seconds": 0.2800006690013106,
-    "gpu_update_seconds": 0.9662619504169725,
-    "producer_batches_per_second": 13.655436069480018,
-    "gpu_batches_per_second": 1.0349160489745752,
-    "producer_headroom": 13.194728290290039,
+    "producer_batch_seconds": 0.09089904831247964,
+    "producer_p95_seconds": 0.2728259089999483,
+    "gpu_update_seconds": 0.9732931058623269,
+    "producer_batches_per_second": 11.00121528844113,
+    "gpu_batches_per_second": 1.0274397239401085,
+    "producer_headroom": 10.707406996346982,
     "sustainable": true,
-    "buffer_seconds": 60.87450287626927,
+    "buffer_seconds": 60.34417256346427,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 418364.321057992,
-    "upload_bytes_per_second": 5309191.789301375,
+    "generation_bytes_per_second": 453332.2705336142,
+    "upload_bytes_per_second": 5312033.959962295,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.77124427609272,
+    "estimated_full_shards_per_hour": 97.27455222135849,
     "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": 1.39643492399955,
-      "bytes": 7665740,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.1344610849992023,
       "bytes": 6680190,
@@ -562,6 +556,12 @@
       "bytes": 5949225,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0684557280001172,
+      "bytes": 6155327,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -232,3 +232,4 @@
 {"step": 743200, "loss": 0.5074306130409241, "lr": 5e-05, "grad_norm": 0.22568622470124966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108784215551, "tokens_per_second": 151115.52954239104, "data_wait_s": 1.3770000805379823e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 743300, "loss": 0.571385383605957, "lr": 5e-05, "grad_norm": 0.2381628199004266, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108798847193, "tokens_per_second": 150964.88080803826, "data_wait_s": 1.134899866883643e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 743400, "loss": 0.5343039631843567, "lr": 5e-05, "grad_norm": 0.23291467367445792, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108813587604, "tokens_per_second": 151039.99727822057, "data_wait_s": 1.371100006508641e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 743500, "loss": 0.5424144268035889, "lr": 5e-05, "grad_norm": 0.24468611795184939, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108828331762, "tokens_per_second": 151985.0482702344, "data_wait_s": 1.322199750575237e-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:db4c32f24db7a76551fa956419b4ef528a14dc6fa4d4079cd6b78347c19256d5
-size 498909
+oid sha256:75467ee40e7e37bb1ec20c3eaf40e0218087a395f2544726de75d51fc42b1288
+size 501027