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": 76492.47235574885,
-      "required_pages_per_second_upper_estimate": 0.005291926724559506
+      "required_token_positions_per_second": 77589.76991885302,
+      "required_pages_per_second_upper_estimate": 0.005367840316056181
     },
     {
       "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": 9561.559044468606,
-      "required_pages_per_second_upper_estimate": 0.004003225103861145
+      "required_token_positions_per_second": 9698.721239856628,
+      "required_pages_per_second_upper_estimate": 0.004060652050797763
     },
     {
       "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": 19123.11808893721,
-      "required_pages_per_second_upper_estimate": 0.003982841187388201
+      "required_token_positions_per_second": 19397.442479713256,
+      "required_pages_per_second_upper_estimate": 0.004039975723566173
     },
     {
       "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": 28684.677133405814,
-      "required_pages_per_second_upper_estimate": 0.0017196025951817383
+      "required_token_positions_per_second": 29096.163719569886,
+      "required_pages_per_second_upper_estimate": 0.0017442705877186372
     },
     {
       "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": 28684.677133405814,
-      "required_pages_per_second_upper_estimate": 0.002230107476561708
+      "required_token_positions_per_second": 29096.163719569886,
+      "required_pages_per_second_upper_estimate": 0.0022620987486977524
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28684.677133405814,
-      "required_pages_per_second_upper_estimate": 0.3679362390606305
+      "required_token_positions_per_second": 29096.163719569886,
+      "required_pages_per_second_upper_estimate": 0.3732143471680697
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06237508400788272,
-    "producer_p95_seconds": 0.20483402399986517,
-    "gpu_update_seconds": 0.9638595502196358,
-    "producer_batches_per_second": 16.032042535984782,
-    "gpu_batches_per_second": 1.0374955560404302,
-    "producer_headroom": 15.452637307836362,
+    "producer_batch_seconds": 0.07721428907031225,
+    "producer_p95_seconds": 0.21163546099887753,
+    "gpu_update_seconds": 0.9502283622842052,
+    "producer_batches_per_second": 12.950970759950767,
+    "gpu_batches_per_second": 1.0523786067552765,
+    "producer_headroom": 12.306379735218645,
     "sustainable": true,
-    "buffer_seconds": 60.72315166383705,
+    "buffer_seconds": 60.81461518618913,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 387152.64690177067,
+    "generation_bytes_per_second": 388665.43624140567,
     "upload_bytes_per_second": 5811189.969732129,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.0739455727562,
+    "estimated_full_shards_per_hour": 83.39855494910839,
     "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": [

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -122,3 +122,4 @@
 {"step": 497200, "loss": 0.5461172461509705, "lr": 0.0003, "grad_norm": 0.1824614696415162, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72773223187, "tokens_per_second": 153278.01444052206, "data_wait_s": 1.3601000318885781e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 497300, "loss": 0.6170888543128967, "lr": 0.0003, "grad_norm": 0.1679577130254991, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72788008628, "tokens_per_second": 154188.58650900476, "data_wait_s": 1.3706001482205465e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 497400, "loss": 0.5712863802909851, "lr": 0.0003, "grad_norm": 0.18592945775216052, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72802695810, "tokens_per_second": 153344.15223244103, "data_wait_s": 1.1845000699395314e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 497500, "loss": 0.6030028462409973, "lr": 0.0003, "grad_norm": 0.16092888357860013, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72817479249, "tokens_per_second": 154625.81571509506, "data_wait_s": 1.7226000636583194e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1b5f5de403a45782d4c4b53d7ff1bd3e835d345a24c4c27233bb56ddcafc1fc8
-size 264363
+oid sha256:73612ab8cb9414ed1e30d38a9dae5498557002636f49e4582a9a1a7c799c389e
+size 266481