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": 76109.47044479584,
-      "required_pages_per_second_upper_estimate": 0.005265429763672906
+      "required_token_positions_per_second": 76546.83873840621,
+      "required_pages_per_second_upper_estimate": 0.0052956879170591295
     },
     {
       "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": 9513.68380559948,
-      "required_pages_per_second_upper_estimate": 0.003983180741095315
+      "required_token_positions_per_second": 9568.354842300776,
+      "required_pages_per_second_upper_estimate": 0.004006070362501078
     },
     {
       "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": 19027.36761119896,
-      "required_pages_per_second_upper_estimate": 0.003962898887985227
+      "required_token_positions_per_second": 19136.709684601552,
+      "required_pages_per_second_upper_estimate": 0.003985671958330602
     },
     {
       "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": 28541.051416798437,
-      "required_pages_per_second_upper_estimate": 0.0017109924527748972
+      "required_token_positions_per_second": 28705.064526902326,
+      "required_pages_per_second_upper_estimate": 0.0017208247882921066
     },
     {
       "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": 28541.051416798437,
-      "required_pages_per_second_upper_estimate": 0.0022189412088382474
+      "required_token_positions_per_second": 28705.064526902326,
+      "required_pages_per_second_upper_estimate": 0.002231692506731395
     },
     {
       "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": 28541.051416798437,
-      "required_pages_per_second_upper_estimate": 0.3660939625812706
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28705.064526902326,
+      "required_pages_per_second_upper_estimate": 0.36819774665412613
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07955510920314168,
-    "producer_p95_seconds": 0.2891347170007066,
-    "gpu_update_seconds": 0.9687099327997142,
-    "producer_batches_per_second": 12.569902926618186,
-    "gpu_batches_per_second": 1.0323007601561933,
-    "producer_headroom": 12.176589819343233,
+    "producer_batch_seconds": 0.13029478168758146,
+    "producer_p95_seconds": 0.2962600310020207,
+    "gpu_update_seconds": 0.9631749816861882,
+    "producer_batches_per_second": 7.674904451643984,
+    "gpu_batches_per_second": 1.0382329472982612,
+    "producer_headroom": 7.392275954655439,
     "sustainable": true,
-    "buffer_seconds": 60.06001583358228,
+    "buffer_seconds": 60.68002384622986,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 157,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 566351.1648585552,
-    "upload_bytes_per_second": 4989871.069013539,
+    "generation_bytes_per_second": 547303.7047509172,
+    "upload_bytes_per_second": 4988861.269562455,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.76288799914118,
+    "estimated_full_shards_per_hour": 117.43863446136127,
     "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": 0.9623302439977124,
-      "bytes": 6111937,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.1251186529989354,
       "bytes": 6276648,
@@ -562,6 +556,12 @@
       "bytes": 5911161,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.0065856220026035,
+      "bytes": 6244496,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -252,3 +252,4 @@
 {"step": 465400, "loss": 0.5856493711471558, "lr": 0.0003, "grad_norm": 0.17388075203956013, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68120467123, "tokens_per_second": 153737.55892069484, "data_wait_s": 1.8632999854162335e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 465500, "loss": 0.6045053601264954, "lr": 0.0003, "grad_norm": 0.17046018346913602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68135156367, "tokens_per_second": 152315.42781032363, "data_wait_s": 1.2090997188352048e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 465600, "loss": 0.6022682785987854, "lr": 0.0003, "grad_norm": 0.1557973367140862, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68149959809, "tokens_per_second": 153728.81772058288, "data_wait_s": 1.483599771745503e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 465700, "loss": 0.592523992061615, "lr": 0.0003, "grad_norm": 0.14624734479052887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68164666775, "tokens_per_second": 152591.0714904238, "data_wait_s": 1.0157000360777602e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4ff7f4ddfbc97c744ff93ccd374e54fcb7a212af81da240f16f912d85c37cb67
-size 542052
+oid sha256:f34bd1a4f68005dbcb2b89cc132cfac08ea101767ab4a8fd01e4599924dba8bf
+size 544170