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": 76807.92495124304,
-      "required_pages_per_second_upper_estimate": 0.005313750467066659
+      "required_token_positions_per_second": 75747.9553695809,
+      "required_pages_per_second_upper_estimate": 0.005240419311938993
     },
     {
       "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": 9600.99061890538,
-      "required_pages_per_second_upper_estimate": 0.004019734280652914
+      "required_token_positions_per_second": 9468.494421197613,
+      "required_pages_per_second_upper_estimate": 0.003964260889507906
     },
     {
       "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": 19201.98123781076,
-      "required_pages_per_second_upper_estimate": 0.003999266301537467
+      "required_token_positions_per_second": 18936.988842395225,
+      "required_pages_per_second_upper_estimate": 0.003944075373891823
     },
     {
       "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": 28802.97185671614,
-      "required_pages_per_second_upper_estimate": 0.0017266941832186084
+      "required_token_positions_per_second": 28405.483263592836,
+      "required_pages_per_second_upper_estimate": 0.0017028653489908057
     },
     {
       "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": 28802.97185671614,
-      "required_pages_per_second_upper_estimate": 0.0022393043709755876
+      "required_token_positions_per_second": 28405.483263592836,
+      "required_pages_per_second_upper_estimate": 0.0022084013812277985
     },
     {
       "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": 28802.97185671614,
-      "required_pages_per_second_upper_estimate": 0.36945359675627737
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28405.483263592836,
+      "required_pages_per_second_upper_estimate": 0.36435503987369117
     }
   ],
   "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": 0.4837688402419076,
-    "producer_p95_seconds": 0.34814255299716024,
-    "gpu_update_seconds": 0.9599009483305512,
-    "producer_batches_per_second": 2.067102956651677,
-    "gpu_batches_per_second": 1.0417741556971984,
-    "producer_headroom": 1.984214088386831,
+    "producer_batch_seconds": 0.11966311317200962,
+    "producer_p95_seconds": 0.34062486000038916,
+    "gpu_update_seconds": 0.9733332027283725,
+    "producer_batches_per_second": 8.356794115514537,
+    "gpu_batches_per_second": 1.0273973981334215,
+    "producer_headroom": 8.133945180995381,
     "sustainable": true,
-    "buffer_seconds": 60.473759744824726,
+    "buffer_seconds": 60.346658569159096,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 254737.91485992292,
-    "upload_bytes_per_second": 3648178.0571288485,
+    "generation_bytes_per_second": 281849.0709739727,
+    "upload_bytes_per_second": 3620895.68875037,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.660826533777865,
+    "estimated_full_shards_per_hour": 60.47824952043901,
     "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": 1.2353659159998642,
-      "bytes": 8486468,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.681274881997524,
       "bytes": 6344982,
@@ -562,6 +556,12 @@
       "bytes": 6867360,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.4055517060041893,
+      "bytes": 5919524,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -388,3 +388,4 @@
 {"step": 881300, "loss": 0.5146641135215759, "lr": 3e-05, "grad_norm": 0.3188050233424422, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128997482835, "tokens_per_second": 151961.0611033549, "data_wait_s": 1.0784999176394194e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 881400, "loss": 0.4751949906349182, "lr": 3e-05, "grad_norm": 0.3020137335532243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129012267463, "tokens_per_second": 152933.44405286823, "data_wait_s": 1.571999746374786e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 881500, "loss": 0.537477433681488, "lr": 3e-05, "grad_norm": 0.31144380978314007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129027006398, "tokens_per_second": 153477.819587664, "data_wait_s": 1.7523998394608498e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 881600, "loss": 0.49659985303878784, "lr": 3e-05, "grad_norm": 0.30103455746506924, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129041759699, "tokens_per_second": 153413.5448973277, "data_wait_s": 1.6828002117108554e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a6445e87ed5fef534bdd1d0ebdce842b06617ab3691f310f14ba0503e391c131
-size 832449
+oid sha256:fbfd1bbb0e0d44dadd2f5e5be14148b4d19c18196f26698f17f510000a48a33f
+size 834567