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": 76606.36439925455,
-      "required_pages_per_second_upper_estimate": 0.00529980604039518
+      "required_token_positions_per_second": 75454.68924719299,
+      "required_pages_per_second_upper_estimate": 0.005220130481121043
     },
     {
       "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": 9575.79554990682,
-      "required_pages_per_second_upper_estimate": 0.004009185631396085
+      "required_token_positions_per_second": 9431.836155899124,
+      "required_pages_per_second_upper_estimate": 0.003948912839339058
     },
     {
       "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": 19151.59109981364,
-      "required_pages_per_second_upper_estimate": 0.003988771364669969
+      "required_token_positions_per_second": 18863.672311798247,
+      "required_pages_per_second_upper_estimate": 0.003928805474055406
     },
     {
       "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": 28727.386649720454,
-      "required_pages_per_second_upper_estimate": 0.0017221629654711455
+      "required_token_positions_per_second": 28295.508467697375,
+      "required_pages_per_second_upper_estimate": 0.0016962725278986631
     },
     {
       "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": 28727.386649720454,
-      "required_pages_per_second_upper_estimate": 0.0022334279535958625
+      "required_token_positions_per_second": 28295.508467697375,
+      "required_pages_per_second_upper_estimate": 0.0021998513245749317
     },
     {
       "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": 28727.386649720454,
-      "required_pages_per_second_upper_estimate": 0.36848407087800894
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28295.508467697375,
+      "required_pages_per_second_upper_estimate": 0.3629444012736801
     }
   ],
   "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": 0.13691381678899006,
-    "producer_p95_seconds": 0.36692739599675406,
-    "gpu_update_seconds": 0.9624265630952908,
-    "producer_batches_per_second": 7.303864748297742,
-    "gpu_batches_per_second": 1.0390403157450976,
-    "producer_headroom": 7.0294334470170465,
+    "producer_batch_seconds": 0.22745344122716915,
+    "producer_p95_seconds": 0.7742767360032303,
+    "gpu_update_seconds": 0.9771162102127771,
+    "producer_batches_per_second": 4.396504157531078,
+    "gpu_batches_per_second": 1.0234197217772487,
+    "producer_headroom": 4.2958954805914855,
     "sustainable": true,
-    "buffer_seconds": 60.63287347500332,
+    "buffer_seconds": 60.58120503319218,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 449074.5555470923,
-    "upload_bytes_per_second": 4430686.986170959,
+    "generation_bytes_per_second": 448745.4376624644,
+    "upload_bytes_per_second": 4472752.546943861,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.3609456997831,
+    "estimated_full_shards_per_hour": 96.2903246632142,
     "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": 1.519131245004246,
-      "bytes": 5792384,
-      "pages": 25,
-      "shards": 2
-    },
     {
       "seconds": 1.7155198749969713,
       "bytes": 12980090,
@@ -562,6 +556,12 @@
       "bytes": 7808426,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.6493455579984584,
+      "bytes": 10530653,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -436,3 +436,4 @@
 {"step": 648000, "loss": 0.5907433032989502, "lr": 0.0001, "grad_norm": 0.16774510031635662, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94850495306, "tokens_per_second": 152423.47781005746, "data_wait_s": 1.1436000932008028e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 648100, "loss": 0.5828615427017212, "lr": 0.0001, "grad_norm": 0.21151722133773335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94865161079, "tokens_per_second": 151659.0039229125, "data_wait_s": 1.2552998668979853e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 648200, "loss": 0.5496563911437988, "lr": 0.0001, "grad_norm": 0.17034659888024165, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94879908023, "tokens_per_second": 151723.80537874435, "data_wait_s": 1.1531003110576421e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 648300, "loss": 0.5251903533935547, "lr": 0.0001, "grad_norm": 0.17222245966699207, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94894560854, "tokens_per_second": 149669.61257868801, "data_wait_s": 1.7400001524947584e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:41df605e4c85937a5ecb14062130e16be6d256b337201a807ef2d5d923a3eca3
-size 934896
+oid sha256:29711d79fbbaa86d019131f44e7dd60fb026207c312ab20f41005740693d406c
+size 937014