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": 76053.68750322475,
-      "required_pages_per_second_upper_estimate": 0.005261570570340761
+      "required_token_positions_per_second": 77269.47468629728,
+      "required_pages_per_second_upper_estimate": 0.005345681548680643
     },
     {
       "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": 9506.710937903094,
-      "required_pages_per_second_upper_estimate": 0.003980261347000873
+      "required_token_positions_per_second": 9658.68433578716,
+      "required_pages_per_second_upper_estimate": 0.004043889435129506
     },
     {
       "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": 19013.421875806187,
-      "required_pages_per_second_upper_estimate": 0.003959994359076687
+      "required_token_positions_per_second": 19317.36867157432,
+      "required_pages_per_second_upper_estimate": 0.0040232984610191576
     },
     {
       "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": 28520.132813709282,
-      "required_pages_per_second_upper_estimate": 0.0017097384144605598
+      "required_token_positions_per_second": 28976.053007361483,
+      "required_pages_per_second_upper_estimate": 0.0017370701339201819
     },
     {
       "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": 28520.132813709282,
-      "required_pages_per_second_upper_estimate": 0.0022173148794592764
+      "required_token_positions_per_second": 28976.053007361483,
+      "required_pages_per_second_upper_estimate": 0.0022527606691347384
     },
     {
       "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": 28520.132813709282,
-      "required_pages_per_second_upper_estimate": 0.36582564120148897
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28976.053007361483,
+      "required_pages_per_second_upper_estimate": 0.3716736959166953
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.08803658298440808,
-    "producer_p95_seconds": 0.1427145930001643,
-    "gpu_update_seconds": 0.9694204504794571,
-    "producer_batches_per_second": 11.358914284270975,
-    "gpu_batches_per_second": 1.0315441555884435,
-    "producer_headroom": 11.011563802415509,
+    "producer_batch_seconds": 0.1806359598359606,
+    "producer_p95_seconds": 0.13853943100002652,
+    "gpu_update_seconds": 0.9541672219116908,
+    "producer_batches_per_second": 5.535996270665716,
+    "gpu_batches_per_second": 1.0480343246296833,
+    "producer_headroom": 5.282266182094587,
     "sustainable": true,
-    "buffer_seconds": 60.10406792972634,
+    "buffer_seconds": 60.11253498043652,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 465061.87184608827,
-    "upload_bytes_per_second": 4404424.170669476,
+    "generation_bytes_per_second": 418527.2958963918,
+    "upload_bytes_per_second": 4897790.454644826,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.79145161187158,
+    "estimated_full_shards_per_hour": 89.80621488255325,
     "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": [
@@ -202,6 +202,12 @@
       "bytes": 1199201,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.0240878049999083,
+      "bytes": 7205273,
+      "pages": 28,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -20,3 +20,4 @@
 {"step": 527000, "loss": 0.6127535104751587, "lr": 0.0003, "grad_norm": 0.17756967529033216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77135404411, "tokens_per_second": 151332.22763061014, "data_wait_s": 1.0801000371429836e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 527100, "loss": 0.49864432215690613, "lr": 0.0003, "grad_norm": 0.2060701343573203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77149910079, "tokens_per_second": 150233.78187562985, "data_wait_s": 9.546999990561744e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 527200, "loss": 0.5773270130157471, "lr": 0.0003, "grad_norm": 0.167763305804286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77164356970, "tokens_per_second": 150880.05545506466, "data_wait_s": 8.96299980013282e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 527300, "loss": 0.5715628266334534, "lr": 0.0003, "grad_norm": 0.15710044105245538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77178990457, "tokens_per_second": 151713.5581075037, "data_wait_s": 1.5919000361463986e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c26f1c12ce6379428172fce8c1b4c867a77de3d2a50b56b6fb6d93f9caf5310b
-size 46761
+oid sha256:abf8b9741e9fb9c1c2ae77e0dbe66c694f01a3c515ef09813f30beb1914dd065
+size 48879