Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74175.58270540083,
-      "required_pages_per_second_upper_estimate": 0.006429351512608522
+      "required_token_positions_per_second": 75205.36431719418,
+      "required_pages_per_second_upper_estimate": 0.006518610372761136
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9271.947838175103,
-      "required_pages_per_second_upper_estimate": 0.00369871862062195
+      "required_token_positions_per_second": 9400.670539649273,
+      "required_pages_per_second_upper_estimate": 0.00375006803081589
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18543.895676350206,
-      "required_pages_per_second_upper_estimate": 0.004289803903306343
+      "required_token_positions_per_second": 18801.341079298545,
+      "required_pages_per_second_upper_estimate": 0.004349359366394098
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27815.84351452531,
-      "required_pages_per_second_upper_estimate": 0.0016938389315823885
+      "required_token_positions_per_second": 28202.01161894782,
+      "required_pages_per_second_upper_estimate": 0.0017173545430742582
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27815.84351452531,
-      "required_pages_per_second_upper_estimate": 0.0021413406329311734
+      "required_token_positions_per_second": 28202.01161894782,
+      "required_pages_per_second_upper_estimate": 0.002171068922591349
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27815.84351452531,
-      "required_pages_per_second_upper_estimate": 0.34537968281069953
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28202.01161894782,
+      "required_pages_per_second_upper_estimate": 0.3501745982461207
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.0840216265156073,
-    "producer_p95_seconds": 0.21519500199974573,
-    "gpu_update_seconds": 0.9939659024024325,
-    "producer_batches_per_second": 11.901697711293968,
-    "gpu_batches_per_second": 1.0060707289686528,
-    "producer_headroom": 11.829881705727276,
+    "producer_batch_seconds": 0.11174089453906788,
+    "producer_p95_seconds": 0.21262714200020127,
+    "gpu_update_seconds": 0.9803555992234397,
+    "producer_batches_per_second": 8.949275053909389,
+    "gpu_batches_per_second": 1.0200380359862493,
+    "producer_headroom": 8.77347190809072,
     "sustainable": true,
-    "buffer_seconds": 60.63192004654839,
+    "buffer_seconds": 60.78204715185326,
     "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": 499683.6338292062,
-    "upload_bytes_per_second": 5252209.694508239,
+    "generation_bytes_per_second": 491188.2842349423,
+    "upload_bytes_per_second": 5194652.380403602,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.22047578007832,
+    "estimated_full_shards_per_hour": 105.39757151876641,
     "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": 1.7666170649999913,
-      "bytes": 10799529,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 0.9170148619996326,
       "bytes": 257577,
@@ -562,6 +556,12 @@
       "bytes": 11055954,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.0256725109993567,
+      "bytes": 2318063,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -125,3 +125,4 @@
 {"step": 207500, "loss": 0.6107757650315762, "lr": 0.0003, "grad_norm": 0.1461868884554233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30370276350, "tokens_per_second": 147719.77099585536, "data_wait_s": 2.2218000594875775e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 207600, "loss": 0.569626696407795, "lr": 0.0003, "grad_norm": 0.14743573959776018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30384928391, "tokens_per_second": 147426.8888508114, "data_wait_s": 1.8900000213761814e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 207700, "loss": 0.5794225446879864, "lr": 0.0003, "grad_norm": 0.14461750445336438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30399624366, "tokens_per_second": 148501.17824133413, "data_wait_s": 1.809899913496338e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 207800, "loss": 0.5990188978612423, "lr": 0.0003, "grad_norm": 0.15050431338140977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30414300144, "tokens_per_second": 147051.9778778555, "data_wait_s": 1.845699989644345e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2d0f3bf3b323a97a7a64c2db045ae1e53941d91b8ed5d1a22a9d82d4216f954d
-size 270717
+oid sha256:5e124cb8c4e003f30502cc2b27c9c2754b38b80bd24efeacfb28c4ef634245a5
+size 272835