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": 77380.99936979407,
-      "required_pages_per_second_upper_estimate": 0.005353397085057869
+      "required_token_positions_per_second": 75984.68117116837,
+      "required_pages_per_second_upper_estimate": 0.005256796552172351
     },
     {
       "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": 9672.624921224258,
-      "required_pages_per_second_upper_estimate": 0.004049726067139491
+      "required_token_positions_per_second": 9498.085146396046,
+      "required_pages_per_second_upper_estimate": 0.003976649908223882
     },
     {
       "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": 19345.249842448517,
-      "required_pages_per_second_upper_estimate": 0.004029105373636332
+      "required_token_positions_per_second": 18996.170292792092,
+      "required_pages_per_second_upper_estimate": 0.003956401309289689
     },
     {
       "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": 29017.874763672775,
-      "required_pages_per_second_upper_estimate": 0.0017395772843529194
+      "required_token_positions_per_second": 28494.255439188135,
+      "required_pages_per_second_upper_estimate": 0.00170818710537048
     },
     {
       "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": 29017.874763672775,
-      "required_pages_per_second_upper_estimate": 0.002256012126733476
+      "required_token_positions_per_second": 28494.255439188135,
+      "required_pages_per_second_upper_estimate": 0.0022153030274127987
     },
     {
       "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": 29017.874763672775,
-      "required_pages_per_second_upper_estimate": 0.3722101405019532
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28494.255439188135,
+      "required_pages_per_second_upper_estimate": 0.3654937140260917
     }
   ],
   "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": 0.08805437375787051,
-    "producer_p95_seconds": 0.2127668730026926,
-    "gpu_update_seconds": 0.952792036810783,
-    "producer_batches_per_second": 11.356619294684583,
-    "gpu_batches_per_second": 1.0495469749592294,
-    "producer_headroom": 10.82049642906716,
+    "producer_batch_seconds": 0.13501641413267862,
+    "producer_p95_seconds": 0.2149449739990814,
+    "gpu_update_seconds": 0.9703008404274961,
+    "producer_batches_per_second": 7.406506878617846,
+    "gpu_batches_per_second": 1.0306081973085988,
+    "producer_headroom": 7.1865398489549275,
     "sustainable": true,
-    "buffer_seconds": 60.025898319079324,
+    "buffer_seconds": 60.15865210650476,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297998.5470893914,
-    "upload_bytes_per_second": 4787096.310507735,
+    "generation_bytes_per_second": 330167.2803034946,
+    "upload_bytes_per_second": 4771732.435892133,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.94355115424448,
+    "estimated_full_shards_per_hour": 70.84621245220784,
     "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": 1.7634025150000525,
-      "bytes": 6258608,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.2659589719987707,
       "bytes": 11586393,
@@ -562,6 +556,12 @@
       "bytes": 6018491,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.925738228001137,
+      "bytes": 863516,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -195,3 +195,4 @@
 {"step": 944500, "loss": 0.577720582485199, "lr": 3e-05, "grad_norm": 0.35585286234893304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138248100173, "tokens_per_second": 150613.3662130495, "data_wait_s": 1.7441001546103507e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 944600, "loss": 0.555435836315155, "lr": 3e-05, "grad_norm": 0.35039679697845094, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138262793802, "tokens_per_second": 153374.08155014302, "data_wait_s": 1.309999788645655e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 944700, "loss": 0.5725507736206055, "lr": 3e-05, "grad_norm": 0.34738436125650035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138277526113, "tokens_per_second": 152562.25768287326, "data_wait_s": 1.4356002793647349e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 944800, "loss": 0.5285707116127014, "lr": 3e-05, "grad_norm": 0.3409574839623096, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138292218106, "tokens_per_second": 153583.49825800018, "data_wait_s": 7.965001714183018e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:55b22316dc292ec2b0867650d0f83e73c0906a1c7491f66852dedd59a713e880
-size 419760
+oid sha256:9d92c60d9014540873ee2903aa06a8b15ab73b10c4746e91adfa5ba6b0f79b53
+size 421878