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": 76812.1644317684,
-      "required_pages_per_second_upper_estimate": 0.005314043764166353
+      "required_token_positions_per_second": 74850.99239548223,
+      "required_pages_per_second_upper_estimate": 0.005178365332150009
     },
     {
       "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": 9601.52055397105,
-      "required_pages_per_second_upper_estimate": 0.004019956153398607
+      "required_token_positions_per_second": 9356.374049435279,
+      "required_pages_per_second_upper_estimate": 0.003917318431190623
     },
     {
       "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": 19203.0411079421,
-      "required_pages_per_second_upper_estimate": 0.00399948704453517
+      "required_token_positions_per_second": 18712.748098870557,
+      "required_pages_per_second_upper_estimate": 0.0038973719406417153
     },
     {
       "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": 28804.561661913147,
-      "required_pages_per_second_upper_estimate": 0.0017267894896126768
+      "required_token_positions_per_second": 28069.122148305836,
+      "required_pages_per_second_upper_estimate": 0.0016827010137230348
     },
     {
       "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": 28804.561661913147,
-      "required_pages_per_second_upper_estimate": 0.002239427971336841
+      "required_token_positions_per_second": 28069.122148305836,
+      "required_pages_per_second_upper_estimate": 0.0021822507840104227
     },
     {
       "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": 28804.561661913147,
-      "required_pages_per_second_upper_estimate": 0.36947398907034473
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28069.122148305836,
+      "required_pages_per_second_upper_estimate": 0.36004056064321693
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.10191386870337737,
-    "producer_p95_seconds": 0.3661101939942455,
-    "gpu_update_seconds": 0.9598479686832933,
-    "producer_batches_per_second": 9.812207236588405,
-    "gpu_batches_per_second": 1.0418316573319282,
-    "producer_headroom": 9.41822718433889,
+    "producer_batch_seconds": 0.1284099832424772,
+    "producer_p95_seconds": 0.3709900179965189,
+    "gpu_update_seconds": 0.984996960500553,
+    "producer_batches_per_second": 7.787556502610043,
+    "gpu_batches_per_second": 1.0152315591835155,
+    "producer_headroom": 7.670719484797209,
     "sustainable": true,
-    "buffer_seconds": 60.470422027047476,
+    "buffer_seconds": 60.08481459053373,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366783.0225817086,
-    "upload_bytes_per_second": 4435833.203760793,
+    "generation_bytes_per_second": 450451.8864905647,
+    "upload_bytes_per_second": 4372409.167800847,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.70309837425656,
+    "estimated_full_shards_per_hour": 96.65648885762887,
     "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": 1.831955179994111,
-      "bytes": 7695941,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.235046139001497,
       "bytes": 923574,
@@ -562,6 +556,12 @@
       "bytes": 7821820,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.7150816520006629,
+      "bytes": 927906,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -434,3 +434,4 @@
 {"step": 647800, "loss": 0.5408211946487427, "lr": 0.0001, "grad_norm": 0.17289111194633108, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94821128922, "tokens_per_second": 149982.78966636682, "data_wait_s": 1.764699845807627e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 647900, "loss": 0.5759742856025696, "lr": 0.0001, "grad_norm": 0.19641977747136147, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94835801434, "tokens_per_second": 152325.6590805752, "data_wait_s": 1.2803000572603196e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 648000, "loss": 0.5907433032989502, "lr": 0.0001, "grad_norm": 0.16774510031635662, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94850495306, "tokens_per_second": 152423.47781005746, "data_wait_s": 1.1436000932008028e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 648100, "loss": 0.5828615427017212, "lr": 0.0001, "grad_norm": 0.21151722133773335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94865161079, "tokens_per_second": 151659.0039229125, "data_wait_s": 1.2552998668979853e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:68ad5169c1b40bb0391b13387c008ca5df007fb3efb815bcaca89ebc0ac024bb
-size 930660
+oid sha256:05a5767308527402d28e90b7bce40b982c78e8f87112d8de29aad2d51f62031e
+size 932778